A method for judging combat intention of unmanned aerial vehicle based on bayesian model and man-machine complementarity

Abstract

The application provides a UAV combat intention judgment method based on a Bayesian model and human-machine complementarity, and the technical scheme is as follows: a human-machine hybrid intelligent framework combining Bayesian inference and human-machine complementarity theory is constructed, real-time situation information of enemy UAVs is collected and analyzed, situation data of the enemy UAVs are subjected to feature extraction, independent intention judgment is carried out by machine intelligence and human intelligence respectively, and relevant parameters are dynamically obtained through Bayesian inference; a confidence feedback mechanism is used to fuse the prediction results of human and machine by combining a human-machine complementarity coefficient; and through dynamic weight distribution and parameter adjustment, more robust and accurate intention recognition results are generated. The application has the beneficial effect that in a complex and dynamic battlefield environment, the complementary advantages of the intuitive judgment of human intelligence and the high-speed calculation of machine intelligence are effectively fused.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) intent prediction technology, and in particular to a method for judging the combat intent of UAVs based on Bayesian models and human-machine complementarity. Background Technology

[0002] In military operations, human intelligence in judgment and decision-making is always a key factor in determining victory or defeat on the battlefield. For example, commanders need to quickly assess enemy behavior and make accurate decisions in the face of complex and ever-changing tactical situations. A human-machine hybrid intelligence approach based on Bayesian models effectively captures the changing trends of enemy drone behavior by introducing a dynamically adjusted reasoning mechanism. Simultaneously, by combining the global intuitive judgment of human intelligence with the rapid computational capabilities of machine intelligence, a highly complementary hybrid decision-making system is formed.

[0003] With their efficient processing of high-dimensional data and precise calculation of complex models, big data and artificial intelligence technologies are gradually penetrating the field of military operations. However, in practical applications, the limitations of AI technology's interpretability and generalization capabilities are becoming increasingly apparent, leading to its inadequacy in addressing the uncertainty, openness, and vulnerability of complex systems. Given the current limitations of AI technology and the system's need for intelligent solutions, researchers have proposed a comprehensive modeling method combining computer simulation and expert experience, giving rise to the concept of human-machine hybrid intelligence. This provides a new approach to UAV intent recognition. Compared to relying solely on machine or human analysis, combining the advantages of human experience and machine learning helps UAVs improve their recognition performance when processing data.

[0004] However, most existing human-machine hybrid intelligence methods combine human and machine intelligence predictions through human-machine interaction mechanisms. This makes it difficult to leverage their complementary strengths and capture real-time data characteristics when dealing with complex and dynamic real-world environments. For example, the human-machine hybrid model proposed in the paper "Bayesian modeling of human–AI complementarity" explores the conditions for achieving complementarity by combining predictions from human and machine classifiers, revealing that the strengths and weaknesses of humans and machines in tasks are complementary. However, similar methods mostly rely on fixed or pre-set parameter models, making it difficult to capture real-time data characteristics and task difficulty when dealing with complex and dynamic real-world environments. This leads to rigid decision-making processes, and traditional human-machine complementary models do not fully consider the uncertainties in prediction tasks. For instance, current methods cannot dynamically feed back and adjust parameters, causing classifiers to perform poorly in high-difficulty tasks or extreme situations, affecting overall decision-making effectiveness.

[0005] To address this issue, this invention proposes a method for determining the combat intent of unmanned aerial vehicles (UAVs) based on Bayesian models and human-machine complementarity, aiming to improve the accuracy of UAV intent recognition. This method combines the probabilistic outputs of human and machine intelligence, dynamically adjusting classification parameters through Bayesian inference to obtain more realistic parameters. Furthermore, to improve the overall decision-making accuracy, we calculate the contribution ratio of human and machine intelligence using a complementarity coefficient, generating a weighted mixed confidence score. Summary of the Invention

[0006] The purpose of this invention is to propose a method for judging the combat intent of unmanned aerial vehicles (UAVs) based on Bayesian models and human-machine complementarity, to assist in UAV combat intent identification and decision optimization. This method combines the intuitive judgment ability of human intelligence with the high-speed computing advantages of machine intelligence, achieving efficient and reliable intent prediction and decision support in complex and uncertain battlefield environments. This invention not only improves the accuracy of battlefield situational awareness and prediction, but also enhances the model's adaptability in dynamic environments by dynamically adjusting decision weights, thereby improving combat effectiveness in high-risk battlefields and reducing the risk of battle losses due to decision-making errors.

[0007] The inventive concept of this invention is as follows: First, based on situational information of both friendly and enemy drones, a complementary model is obtained by fusing prediction results from human and machine intelligence using methods such as Bayesian inference. Then, by analyzing the uncertainty distribution in the prediction results, a human-machine complementarity coefficient is calculated, and based on this, the combined weights of human and machine intelligence are dynamically adjusted to generate a more accurate final prediction result. Furthermore, a dynamic feedback mechanism is used to continuously optimize the model parameters to adapt to real-time changes in the battlefield environment. This invention endows drones with the ability to identify the intentions of enemy drones in combat and enhances decision-making reliability by quantifying uncertainty and complementarity, thereby providing new technical support and practical application value for the field of military intelligence.

[0008] This invention is achieved through the following measures: a method for determining the combat intent of a drone based on a Bayesian model and human-machine complementarity, comprising the following steps:

[0009] S1. Utilize the sensor system of our UAV to obtain the situational information of the enemy UAV, and then organize the data based on the situational information of our UAV as training data. Use this data to obtain the prediction results of human intelligence and machine intelligence.

[0010] S2. Based on the prediction results, a complementary model is constructed using methods such as Bayesian inference, logit score and confidence, and human-machine complementarity coefficient.

[0011] S3. Through dynamic combination of weight allocation and parameter update mechanism, human-machine hybrid decision-making is realized to obtain the final intention of the enemy drone, providing reliable intelligence support for decision-makers.

[0012] Furthermore, step one includes the following steps:

[0013] 1-1) Collect current status information of enemy drones: position information, flight speed, heading angle, roll angle, pitch angle, control variables, etc., in order to predict the future behavior of enemy drones.

[0014] 1-2) Obtain the situational information of both the red and blue UAVs. The situational information of the red UAV is recorded as follows: The situational information of the blue team's drone is recorded as follows: Among them, x r ,x b ,y r ,y b ,z r ,z b These are the location information of the drone, γ r ,γ b The pitch angle, For yaw angle, nx r ,nx b ,nz r ,nz b To control the amount, roll r ,roll b For roll angle;

[0015] 1-3) Let M be the independently completed machine intelligence training model, and its output result be: in For model M, the result of judging data x is category y. i The probability of x is given, and the total number of categories is k. Let H be the independently trained human intelligence model, and Y be the human's category judgment for x. H (x), with a confidence level of C. H Based on the principle of probability allocation, the human performance on x in each category can be represented by the function: in For model H, the result of judging data x is category y. i The probability of f is given by the total number of categories k. yc This is a probability allocation model built based on expert knowledge.

[0016] Furthermore, step two includes the following steps:

[0017] 2-1) Construct the prior distribution of each parameter, μ H1 ~N(1,2),μ H0 ~N(-1,2), μ M1 ~N(1,2),μ M0 ~N(-1,2), σH ~N(1,2),σ M ~N(1,2),ρ HM ~N(0.5,0.3), where μ H1 Let μ be the mean of human intelligence on positive examples. H0 Let μ be the mean of human intelligence on negative examples. M1 Let μ be the mean of machine intelligence on positive examples. M0 Let σ be the mean of machine intelligence on negative examples. H Let σ be the standard deviation of human intelligence. M ρ represents the standard deviation of machine intelligence. HM The correlation coefficient between human intelligence and machine intelligence;

[0018] 2-2) Calculate the mean, standard deviation, and correlation for each label. The mean for positive human intelligence observations is: The mean of the negative examples is: The variance is: The correlation coefficient is: Where ph[j] represents the probability output of human intelligence on label j, correct and incorrect represent the correctness of each sample in class j, num_correct and num_incorrect record the number of samples correctly and incorrectly classified in this class, Cov(ph[j], pm[j]) represents the covariance between human intelligence and machine intelligence, and σ H [j] represents the variance of human intelligence on label j, σ M [j] represents the variance of machine intelligence on label j;

[0019] 2-3) Construct a logit score based on each data sample i and each class label j to reflect the classifier's performance under different class labels.

[0020] The mean of the logit score is generated by the true label z. i The matching result determines the outcome. During the generation process, when label j matches the real label, the positive class mean is used. and Otherwise, use the negative mean. and Specifically, it is expressed as follows:

[0021]

[0022] Construct a covariance matrix ∑ to represent the potential correlation between the logit scores of human and machine classifiers on different labels;

[0023]

[0024] in, and Let represent the variance of the logit scores of the human and machine classifiers, respectively, and ρ represent the variance of the logit scores of the human and machine classifiers, respectively. HM [j] indicates the correlation between the two.

[0025] Construct human and machine logit scores, sampling the logit scores using a bivariate normal distribution, as follows:

[0026]

[0027] The confidence score is constructed by performing a non-linear transformation on the logit score using the sigmoid function. The corresponding confidence score calculation formula is as follows: Where, γ H [i,j] and γ M [i,j] represent the confidence scores of the human and machine classifiers on label j, respectively;

[0028] 2-4) Constructing the human-machine complementarity coefficient r HM The human-machine complementarity coefficient is a key indicator used to measure the effectiveness of collaboration between humans and machines in hybrid intelligent systems. It describes the degree and effectiveness of the integration of human and machine intelligence in a given task. The formula for calculating the human-machine complementarity coefficient is as follows:

[0029]

[0030] Where, μ H μ represents the mean of human intelligence. M σ represents the mean of machine intelligence. H σ represents the variance of human intelligence. M ρ represents the variance of machine intelligence. HM The correlation coefficient between human intelligence and machine intelligence.

[0031] Furthermore, step three includes the following steps:

[0032] 3-1) Construct a human-machine hybrid intelligent model. Through dynamic weight allocation, the system can flexibly adjust the weights of humans and machines in different tasks, making the predictions of the human-machine combination more accurate. The weight allocation uses parameters α and β to represent the weights of humans and machines in decision-making. A complementarity coefficient is introduced to balance the confidence levels of the human classifier and the machine classifier. The specific calculation formula is as follows:

[0033]

[0034] β=1-α (6)

[0035] The final combined score is: combined_scores = α·γ H +β·γ M , where r HM Human-machine complementarity coefficient, γ H γ represents the confidence score for human intelligence. M To limit the value of the clip function for the confidence score of machine intelligence to the range [0.1, 0.9], we prevent the parameter from going out of bounds.

[0036] 3-2) Construct a dynamic update mechanism to adjust parameters in real time, adapting to changes in human and machine performance during tasks. This mechanism reduces the error rate by continuously adjusting key parameters, thereby improving system stability. The relevant parameter ρ... HM The update is based on the difference between human and machine confidence levels. The formula is: ρ HM =clip(ρ HM +0.001·sign(confidence_diff-0.1),-1.0,1.0), where confidence_diff is the absolute difference between human and machine confidence levels, and the sign function is used to indicate whether to increase or decrease ρ. HM The clip function restricts the value to the range [-1.0, 1.0] to prevent parameters from going out of bounds.

[0037] The variance parameter is updated based on the error rate of each intelligence. When the error rate is high, the corresponding variance parameter will increase, indicating increased uncertainty; conversely, the variance parameter will decrease. The update formula is: σ H =clip(σ H +0.001·sign(errors H -0.5), 1e-6, 10.0), σ M =clip(σ M +0.001·sign(errors M -0.5), 1e-6, 10.0).

[0038] Among them, errors H and errors M These represent the average error rates for humans and machines, respectively.

[0039] 3-3) Use the training set data to train the model, and save the trained model and parameters such as accuracy.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] (1) This invention dynamically quantifies the complementarity coefficients of human intelligence and machine intelligence using a Bayesian model, enabling efficient data fusion and real-time decision optimization in complex and dynamic battlefield environments. Through a mechanism of dynamically updating parameters, this invention effectively captures uncertainties and environmental changes in prediction results, making decision-making more flexible and reliable, especially suitable for high-risk, high-dynamic scenarios in UAV operations.

[0042] (2) Based on the situational information of enemy and friendly UAVs, this invention combines the advantages of both human and machine intelligence to construct a Bayesian inference model and a multi-layered hybrid intelligent architecture. By modeling the combat intentions of enemy UAVs and combining relative position and dynamic situational information, the accuracy of intention recognition and decision-making efficiency are significantly improved, providing more reliable technical support for battlefield situational awareness and action decision-making.

[0043] (3) This invention overcomes the limitations of traditional systems in highly uncertain environments by introducing human-machine complementarity coefficients and dynamic weight adjustment mechanisms. It not only improves the overall accuracy of intent recognition, but also optimizes parameter updates through dynamic feedback mechanisms, enabling the system to have stronger robustness and generalization ability in long-term operation, providing a brand-new solution for UAV combat and intelligent decision-making.

[0044] Figure and Table Description

[0045] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used to explain the invention in conjunction with Embodiment 1 and do not constitute a limitation thereof.

[0046] Figure 1 The overall flowchart of the UAV combat intent judgment method based on Bayesian model and human-machine complementarity provided by the present invention.

[0047] Figure 2 The model framework diagram of the UAV combat intent judgment method based on Bayesian model and human-machine complementarity provided by the present invention is shown.

[0048] Figure 3 This is a schematic diagram illustrating the connections between neurons in the machine intelligent neural network provided by the present invention.

[0049] Figure 4 This provides a schematic diagram of the K-fold cross-validation results for each model in this invention.

[0050] Figure 5 This diagram illustrates the accuracy of the four models provided by this invention under different noise levels. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] Example 1

[0053] This embodiment 1 provides a method for determining the combat intent of a UAV based on a Bayesian model and human-machine complementarity, which includes the following steps:

[0054] Step 1) Use the sensor system of our drone to obtain the situational information of the enemy drone, and then organize the data based on the situational information of our drone as training data. Use this data to obtain the prediction results of human intelligence and machine intelligence.

[0055] Step 2) Based on the prediction results, a complementary model is constructed using methods such as Bayesian inference, logit score and confidence, and human-machine complementarity coefficient;

[0056] Step 3) Achieve human-machine hybrid decision-making through dynamic combination weight allocation and parameter update mechanism to obtain the final enemy drone intentions and provide reliable intelligence support for decision-makers;

[0057] Step 4) Compare the models and analyze the accuracy, F1 score, recall, and precision of each model.

[0058] Step 1) Obtain the situational information of the enemy drone, and then organize the data based on the situational information of our own drone as training data. Use this data to obtain the prediction results of human intelligence and machine intelligence. The specific steps are as follows:

[0059] 1-1) Obtain the situational information of both the red and blue UAVs. The situational information of the red UAV is recorded as follows: The situational information of the blue team's drone is recorded as follows: Among them, x r ,x b ,y r ,y b ,z r ,z b These are the location information of the drone, γ r ,γ b The pitch angle, For yaw angle, nx r ,nx b ,nz r ,nz b To control the amount, roll r ,roll b This is the roll angle.

[0060] 1-2) Let M be the independently completed machine intelligence training model, and its output result be: in For model M, the result of judging data x is category y. i The probability of x is given, and the total number of categories is k. Let H be the independently trained human intelligence model, and Y be the human's category judgment for x. H (x), with a confidence level of C. H Based on the principle of probability allocation, the human performance on x in each category can be represented by the function: in For model H, the result of judging data x is category y. i The probability of f is given by the total number of categories k. yc This is a probability allocation model built based on expert knowledge.

[0061] Step 2) Based on the prediction results, a complementary model is constructed using methods such as Bayesian inference, logit score and confidence level, and human-machine complementarity coefficient. The specific steps are as follows:

[0062] 2-1) Construct the prior distribution of each parameter, μ H1 ~N(1,2),μ H0 ~N(-1,2), μ M1 ~N(1,2),μ M0 ~N(-1,2), σ H ~N(1,2),σ M ~N(1,2),ρ HM ~N(0.5,0.3), where μ H1 Let μ be the mean of human intelligence on positive examples. H0 Let μ be the mean of human intelligence on negative examples. M1 Let μ be the mean of machine intelligence on positive examples. M0 Let σ be the mean of machine intelligence on negative examples. H Let σ be the standard deviation of human intelligence. M ρ represents the standard deviation of machine intelligence. HM The correlation coefficient between human intelligence and machine intelligence;

[0063] 2-2) Calculate the mean, standard deviation, and correlation for each label. The mean for positive human intelligence observations is:

[0064] The mean of the negative examples is: The variance is: The correlation coefficient is: Where ph[j] represents the probability output of human intelligence on label j, correct and incorrect represent the correctness of each sample in class j, num_correct and num_incorrect record the number of samples correctly and incorrectly classified in this class, Cov(ph[j], pm[j]) represents the covariance between human intelligence and machine intelligence, and σ H [j] represents the variance of human intelligence on label j, σ M [j] represents the variance of machine intelligence on label j;

[0065] 2-3) Construct a logit score based on each data sample i and each class label j to reflect the classifier's performance under different class labels.

[0066] The mean of the logit score is generated by the true label z. i The matching result determines the outcome. During the generation process, when label j matches the real label, the positive class mean is used. and Otherwise, use the negative mean. and Specifically, it is expressed as follows:

[0067]

[0068] Construct a covariance matrix ∑ to represent the potential correlation between the logit scores of human and machine classifiers on different labels;

[0069]

[0070] in, and Let represent the variance of the logit scores of the human and machine classifiers, respectively, and ρ represent the variance of the logit scores of the human and machine classifiers, respectively. HM [j] indicates the correlation between the two.

[0071] Construct human and machine logit scores, sampling the logit scores using a bivariate normal distribution, as follows:

[0072]

[0073] The confidence score is constructed by performing a non-linear transformation on the logit score using the sigmoid function. The corresponding confidence score calculation formula is as follows: Where, γ H [i,j] and γ M [i,j] represent the confidence scores of the human and machine classifiers on label j, respectively;

[0074] 2-4) Constructing the human-machine complementarity coefficient rHM The human-machine complementarity coefficient is a key indicator used to measure the effectiveness of collaboration between humans and machines in hybrid intelligent systems. It describes the degree and effectiveness of the integration of human and machine intelligence in a given task. The formula for calculating the human-machine complementarity coefficient is as follows:

[0075]

[0076] Where, μ H μ represents the mean of human intelligence. M σ represents the mean of machine intelligence. H σ represents the variance of human intelligence. M ρ represents the variance of machine intelligence. HM The correlation coefficient between human intelligence and machine intelligence.

[0077] Step 3) Achieve hybrid human-machine decision-making through dynamic weight allocation and parameter update mechanisms to obtain the final enemy drone intent. The specific steps are as follows:

[0078] 3-1) Construct a human-machine hybrid intelligent model. Through dynamic weight allocation, the system can flexibly adjust the weights of humans and machines in different tasks, making the predictions of the human-machine combination more accurate. The weight allocation uses parameters α and β to represent the weights of humans and machines in decision-making. A complementarity coefficient is introduced to balance the confidence levels of the human classifier and the machine classifier. The specific calculation formula is as follows:

[0079]

[0080] β=1-α (6)

[0081] The final combined score is: combined_scores = α·γ H +β·γ M , where r HM Human-machine complementarity coefficient, γ H γ represents the confidence score for human intelligence. M To limit the value of the clip function for the confidence score of machine intelligence to the range [0.1, 0.9], we prevent the parameter from going out of bounds.

[0082] 3-2) Construct a dynamic update mechanism to adjust parameters in real time, adapting to changes in human and machine performance during tasks. This mechanism reduces the error rate by continuously adjusting key parameters, thereby improving system stability. The relevant parameter ρ... HM The update is based on the difference between human and machine confidence levels. The formula is: ρ HM ·clip(ρ HM+0.001·sign(confidence_diff-0.1),-1.0,1.0), where confidence_diff is the absolute difference between human and machine confidence levels, and the sign function is used to indicate whether to increase or decrease ρ. HM The clip function restricts the value to the range [-1.0, 1.0] to prevent parameters from going out of bounds.

[0083] The variance parameter is updated based on the error rate of each intelligence. When the error rate is high, the corresponding variance parameter will increase, indicating increased uncertainty; conversely, the variance parameter will decrease. The update formula is: σ H =clip(σ H +0.001·sign(errors H -0.5), 1e-6, 10.0), σ M =clip(σ M +0.001·sign(errors M -0.5), 1e-6, 10.0).

[0084] Among them, errors H and errors M These represent the average error rates for humans and machines, respectively.

[0085] 3-3) Use the training set data to train the model, and save the trained model and parameters such as accuracy.

[0086] Step 4) Compare the models and analyze their accuracy, F1 score, recall, and precision. The specific steps are as follows:

[0087] Table 1 compares the accuracy, precision, recall, and F1 score of the four models across different categories. The B-HM-C model, after incorporating human-machine hybrid optimization, generally outperforms the other three models, demonstrating superior performance.

[0088] Table 1 shows a comparison of the four models.

[0089]

[0090] Based on the above comparison results, it can be seen that the UAV combat intent judgment method proposed in this invention, based on Bayesian models and human-machine complementarity, effectively combines the advantages of both humans and machines. Unlike static Bayesian models, this model, based on confidence feedback and dynamic parameter update mechanisms, demonstrates strong accuracy, providing effective support for UAV combat decision-making in complex battlefield environments.

[0091] Example 2

[0092] This embodiment 2 provides a method for determining the combat intent of a UAV based on a Bayesian model and human-machine complementarity, which includes the following steps:

[0093] Step 1) Use the sensor system of our drone to obtain the situational information of the enemy drone, and then organize the data based on the situational information of our drone as training data. Use this data to obtain the prediction results of human intelligence and machine intelligence.

[0094] Step 2) Based on the prediction results, a complementary model is constructed using methods such as Bayesian inference, logit score and confidence, and human-machine complementarity coefficient;

[0095] Step 3) Achieve human-machine hybrid decision-making through dynamic combination weight allocation and parameter update mechanism to obtain the final enemy drone intentions and provide reliable intelligence support for decision-makers;

[0096] Step 4) Compare the models and analyze the K-fold cross-validation of each model.

[0097] Steps 1), 2), and 3) are the same as those in Example 1 above.

[0098] Step 4) By evaluating the model's performance on unseen data, ensure the model has good generalization ability, reduce the risk of overfitting and underfitting, and thus build a more reliable and robust model. Figure 4 The results of K-fold cross-validation for each model are presented. Figure 4 The results show that the UAV combat intent judgment method based on Bayesian model and human-machine complementarity proposed in this invention performs significantly better than other models in UAV intent judgment tasks, proving that the method has high generalization ability, can effectively overcome the limitations of traditional methods, adapt to the ever-changing battlefield environment, and provide more reliable support for UAV air combat decision-making.

[0099] Example 3

[0100] This embodiment 3 provides a method for determining the combat intent of a UAV based on a Bayesian model and human-machine complementarity, which includes the following steps:

[0101] Step 1) Use the sensor system of our drone to obtain the situational information of the enemy drone, and then organize the data based on the situational information of our drone as training data. Use this data to obtain the prediction results of human intelligence and machine intelligence.

[0102] Step 2) Based on the prediction results, a complementary model is constructed using methods such as Bayesian inference, logit score and confidence, and human-machine complementarity coefficient;

[0103] Step 3) Achieve human-machine hybrid decision-making through dynamic combination weight allocation and parameter update mechanism to obtain the final enemy drone intentions and provide reliable intelligence support for decision-makers;

[0104] Step 4) Compare the models and analyze the robustness and stability of each model.

[0105] Steps 1), 2), and 3) are the same as those in Example 1 above.

[0106] Step 4) Test the robustness and stability of the model by setting noise levels at 5%, 10%, 15%, 20%, and 25% of the data. Figure 5 The accuracy of each model under different noise levels is shown. Figure 5 The results demonstrate the superiority of the proposed UAV combat intent judgment method based on Bayesian model and human-machine complementarity across all noise levels, exhibiting the smallest decrease in accuracy under different noise conditions. The accuracy drops from 0.928 at 5% noise to 0.713 at 25% noise. Even in high-noise environments, the model maintains a robust and high accuracy, better addressing uncertainties in complex environments and providing more reliable and robust support for UAV air combat intent recognition.

[0107] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1.A method for judging combat intention of a UAV based on a Bayesian model and human-machine complementarity, characterized in that, It comprises the following steps: 1) both sides of the unmanned aircraft situation information acquisition, human intelligence and machine intelligence prediction; 2) based on the prediction results through Bayesian inference, logit score and confidence, human-machine complementary coefficient method constitutes a complementary model; In the step 2), based on the prediction results through Bayesian inference, logit score and confidence, human-machine complementary coefficient method constitutes a complementary model, comprising the following steps: S31, constructing the prior distribution of each parameter, , , , , , , wherein represents a normal distribution, is the mean of human intelligence on positive examples, is the mean of human intelligence on negative examples, is the mean of machine intelligence on positive examples, is the mean of machine intelligence on negative examples, is the standard deviation of human intelligence, is the standard deviation of machine intelligence, is the correlation coefficient of human intelligence and machine intelligence; S32, calculate the mean, standard deviation and correlation of each label, the mean of human intelligence positive observation is: , the mean of the negative observations is: , the variance is: , the correlation coefficient is: where, represents the probability output of human intelligence on label, correct and incorrect represent the correctness of each sample on class j, num_correct and num_incorrect record the number of samples correctly and incorrectly classified on this class, represents the covariance of human intelligence and machine intelligence, represents the variance of human intelligence on label, represents the variance of machine intelligence on label; S33, construct the logit score based on each data sample i and each category label j, which reflects the performance of the classifier under different category labels; The mean of the logit score generation is determined by the matching of the true label In the generation process, when the label j matches the true label, the positive class mean and is used; otherwise, the negative class mean and is used, denoted as: (1)； Constructing a covariance matrix to represent potential correlations between logit scores of human and machine intelligence on different labels; (2)； where, and denote the variance of the logit scores for the human and machine classifiers, respectively, while denotes the correlation between the two, constructing the logit scores, sampling from the bivariate normal distribution of the logit scores, denoted as: (3)； The confidence score is constructed, and the logit score is nonlinearly transformed by a sigmoid function. The corresponding confidence score calculation formula is: , , and respectively represent the confidence scores of the human and machine classifiers on label j. S34、Constructing the human-machine complementary coefficient It is a key indicator for measuring the collaboration effect of human and machine in a hybrid intelligent system, describing the degree of combination and effectiveness of human intelligence and machine intelligence in a given task. The calculation formula of the human-machine complementary coefficient is as follows: (4)； wherein, represents the mean of human intelligence, represents the mean of machine intelligence, represents the variance of human intelligence, represents the variance of machine intelligence, is the correlation coefficient of human intelligence and machine intelligence; 3) realize human-machine hybrid decision through dynamic combination weight distribution and parameter updating mechanism, and get the final enemy unmanned aircraft intention. 2.The method according to claim 1, wherein, In the step 1), the acquisition of both sides of the unmanned aircraft situation information, human intelligence and machine intelligence prediction comprises the following steps: S21, acquire the situation information of the red and blue unmanned aerial vehicles, the situation information of the red unmanned aerial vehicle is denoted as: , and the situation information of the blue unmanned aerial vehicle is denoted as: , wherein, represents position information of the red unmanned aerial vehicle in three-dimensional space, represents a speed of the red unmanned aerial vehicle, represents a pitch angle of the red unmanned aerial vehicle, represents a heading angle of the red unmanned aerial vehicle, represents a control amount of the red unmanned aerial vehicle, represents a roll angle of the red unmanned aerial vehicle, represents position information of the blue unmanned aerial vehicle in three-dimensional space, represents a speed of the blue unmanned aerial vehicle, represents a pitch angle of the blue unmanned aerial vehicle, represents a heading angle of the blue unmanned aerial vehicle, represents a control amount of the blue unmanned aerial vehicle, represents a roll angle of the blue unmanned aerial vehicle; S22, set the independently completed machine intelligence training model as , and the output result is: , wherein is the model , the judgment result of the data is the category , and the total category is , set the independently completed human intelligence training model as , the category judgment result of the human to is , the credibility is , based on the probability distribution principle, the performance of the human to in each category is expressed using a function: , wherein is the model , the judgment result of the data is the category , and the total category is , is a probability distribution model constructed based on expert knowledge. 3.The method according to claim 1, wherein, In the step 3), realize human-machine hybrid decision through dynamic combination weight distribution and parameter updating mechanism, and get the final enemy unmanned aircraft intention, comprising the following steps: S41, build a human-machine hybrid intelligent model, dynamically combine the weight distribution, and flexibly adjust the weight of human and machine in different tasks. The combination weight distribution adopts parameters α and β to represent the weight of human and machine in decision-making. A complementary coefficient is introduced to balance the confidence of human classifier and the confidence of machine classifier. The calculation formula is as follows: (5)； (6)； The final score after combination is: where, is the human-machine complementarity coefficient, is the confidence score of human intelligence, is the confidence score of machine intelligence clip function limits the value between [0.1, 0.9] to avoid parameter out-of-bound; S42, constructing a dynamic updating mechanism to adjust parameters in real time, correlation parameters The update of the correlation parameter is based on the difference between the human and machine confidence, and the formula is: , wherein is the absolute difference between the human and machine confidence, the sign function is used to indicate the direction of increase or decrease , and the clip function limits the value to [−1.0, 1.0] to avoid parameter out-of-bounds; the update of the variance parameter is based on the error rate of each intelligence, and when the error rate is high, the corresponding variance parameter will increase, indicating that the uncertainty is improved; otherwise, the variance parameter decreases, and the formula is: , ; where, and respectively represent the average error rates of humans and machines; S43, use the training set data to train the model, save the trained model and the accuracy parameter.

Images