A multi-core correlation entropy extended Kalman filter dual-band infrared radiation attitude estimation method

Abstract

The application discloses a dual-band infrared radiation attitude estimation method based on multi-core related entropy extended Kalman filtering, relates to the technical field of attitude information acquisition and processing of rotary aircrafts, and comprises the following steps: collecting infrared radiation attitude data by using a semi-physical experiment platform, and analyzing the noise characteristics of dual-band infrared radiation data in an actual application scene; constructing a multi-core generalized related entropy induced function extended Kalman filtering estimation algorithm, selecting Cauchy function and Gaussian function as kernel functions, and integrating an adaptive kernel size mechanism into the update step of the extended Kalman filter; performing the update step and the prediction step of the extended Kalman filter on the collected dual-band infrared radiation data, iteratively optimizing the dual-band infrared radiation attitude estimation under strong interference by using an improved particle swarm optimization algorithm to obtain the best parameter setting; and the method provides a robust attitude estimation method for rotary aircrafts using infrared radiation as an attitude measurement method.

Description

Technical Field

[0001] This invention relates to the field of attitude information acquisition and processing technology for rotating aircraft, and in particular to a dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering. Background Technology

[0002] Currently, mainstream attitude measurement technologies include inertial navigation systems, geomagnetic systems, and GPS systems. However, these technologies suffer from drawbacks such as accumulated errors, high cost, large size, high energy consumption, and susceptibility to electromagnetic interference. To overcome these shortcomings, researchers have proposed dual-band infrared radiation attitude measurement technology. This technology utilizes infrared radiation sensors to extract navigation information from the infrared fields of different regions of the Earth. Compared to mainstream attitude measurement technologies, it offers advantages such as lower cost, lower energy consumption, and smaller size.

[0003] Dual-band infrared attitude measurement is commonly used in rotating aircraft, which are complex nonlinear systems characterized by multiple variables and strong coupling. Attitude information is easily affected by noise, thus reducing its reliability. The main source of noise in dual-band infrared attitude measurement is mechanical noise, including sensor measurement noise, assembly position, and motor interference. Mechanical noise is common to all attitude measurement technologies and affects the robustness of dual-band infrared attitude measurement.

[0004] Regarding improving the reliability of attitude measurement, some researchers have mitigated the impact of measurement noise on the reliability of attitude estimation through infrared sensor manufacturing technology. However, the transition from the development of new sensor technologies to their commercial application still requires a long period of transformation. Other researchers have used statistical methods to study the infrared radiation noise characteristics in cloud, mountain, and snow environments, and have theoretically achieved noise compensation. However, these methods are limited in practical applications due to the difficulty in obtaining large amounts of data in advance to extract statistical features.

[0005] Existing technologies employ Kalman filters to mitigate the impact of noise on aircraft attitude measurement accuracy. However, Kalman filters are only applicable to linear Gaussian systems, while infrared radiation attitude measurement systems are clearly nonlinear. Interactive multi-model extended Kalman filters aim to reduce the noise impact in infrared radiation attitude information to improve information reliability. Adaptive fault-tolerant extended Kalman filters effectively filter out mechanical noise in specific scenarios, enhancing system robustness. Interactive multi-model capacitive Kalman filters are used to filter out noise during infrared attitude measurement, improving the accuracy of aircraft attitude estimation. These methods are essentially nonlinear Gaussian filters based on mean square error. However, in real-world scenarios, noise (mechanical noise or radiated noise) is often non-Gaussian or has a non-zero mean Gaussian distribution. This noise poses a challenge to traditional filtering methods based on minimizing mean square error, as these methods are sensitive to outliers. Summary of the Invention

[0006] The problem to be solved by this invention is to provide a dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering, which can be used to achieve high-precision and robust dual-band infrared radiation attitude measurement.

[0007] This invention adopts the following technical solution: a dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering, comprising the following steps:

[0008] S11. Collect dual-band infrared radiation data under different radiation backgrounds through a semi-physical experimental platform and perform data preprocessing; by recording the roll angle and pitch angle parameter information set by the rotating aircraft, deduce the theoretical dual-band infrared radiation data, remove the theoretical data from the actual collected data, and obtain pure noise data.

[0009] S12. Analyze the probability distribution characteristics of the actual measured dual-band infrared radiation noise, and use the D'Agostino-Pearson normality test algorithm to judge the pure noise data.

[0010] S21. Construct a multi-kernel generalized correlation entropy induced function extended Kalman filter estimation algorithm, and select the Laplace kernel and Cauchy kernel as the kernel functions of the multi-kernel correlation entropy extended Kalman filter;

[0011] S22, The multi-core correlation entropy extended Kalman filter selects an adaptive kernel size mechanism to get rid of the fixed kernel size limitation;

[0012] S23. The optimal parameters of the multi-core correlated entropy extended Kalman filter are replaced by improving the particle swarm optimization algorithm. The linear differential decreasing strategy of inertial weight is introduced to adjust the inertial weight in different search stages. Combined with the simulated annealing strategy, the ability of particles to jump out of local optima is enhanced.

[0013] S24. Input the collected dual-band infrared radiation data into the extended Kalman filter with the multi-core generalized correlation entropy induced function under the optimal parameters to obtain the attitude information of the rotating aircraft.

[0014] Preferably, in step S11, firstly, the sampling rate and sampling time interval of the hardware-in-the-loop experimental platform are set, dual-band infrared radiation data under different radiation backgrounds are collected, the collected data are normalized, the dual-band infrared radiation data are unified into the (-1,1) interval, the data is cleaned, and unusable data is deleted.

[0015] Then, by controlling the attitude change of the rotating aircraft through the motor, the set roll angle and pitch angle parameter information are recorded, and the theoretical dual-band infrared radiation data are derived from the roll angle and pitch angle parameter information.

[0016] Finally, theoretical data was removed from the actual measurement data, and the actual acquired dual-band data was used. And theoretically interference-free data By performing differential analysis, pure noise data of dual-band infrared radiation can be obtained.

[0017] Preferably, in step S12, the dual-band infrared radiation attitude noise is caused by factors such as sensor measurement noise, assembly position, and motor interference. From a time series perspective, the pure noise data has the characteristics of numerical randomness, amplitude limitation, and non-temporal order. Its impact on the dual-band infrared radiation attitude information is that it superimposes a random and... Amplitude-limited noise.

[0018] The D'Agostino-Pearson normality test algorithm was used to judge the pure noise data. The skewness and kurtosis of the noise were calculated and standardized. The D'Agostino statistic was calculated using the skewness and kurtosis. The calculation results were verified by comparing them with the chi-square distribution table. The results showed that the dual-band infrared radiation noise was non-Gaussian distributed, and the noise reduced the reliability of the dual-band infrared radiation attitude.

[0019] Preferably, in step S21, the type of kernel function is crucial for the estimation algorithm based on maximum correlation entropy. The kernel function maps the original error space to the regenerating kernel Hilbert space through Mercer's theorem, making the maximum correlation entropy criterion equivalent to achieving nonlinear separation of signal features in the regenerating kernel Hilbert space, which can significantly improve robustness under non-Gaussian noise. The Laplace kernel and Cauchy kernel are chosen as the kernel functions of the multi-kernel correlation entropy extended Kalman filter; they have good performance against impulse noise and low sensitivity to parameters.

[0020] Preferably, in step S22, the choice of kernel size has a significant impact on the estimation algorithm based on maximum correlation entropy.

[0021] However, engineering practice typically employs fixed-size cores, selected through trial and error based on experience for specific types of non-Gaussian noise. But noise is unstable; initial strong noise tends to plateau over time, and using a fixed core size can hinder the MCC filter from achieving optimal performance.

[0022] Therefore, an adaptive kernel size mechanism is chosen to control the kernel size of the multi-kernel correlation entropy extended Kalman filter, thus breaking free from the limitation of a fixed kernel size.

[0023] Preferably, in step S23, parameters need to be preset during the selection of kernel function type and size. The same applies to the extended Kalman filter. An inertial weight linear differential decreasing strategy is introduced to adjust the inertial weight at different search stages. By setting the inertial weight linear differential decreasing strategy, the particle velocity v and particle position x are updated.

[0024] Preferably, in step S23, after the particle's position is updated, a decision is made based on the Metropolis criterion of simulated annealing to determine whether to accept the new solution, thereby enhancing the particle's ability to escape local optima. The hyperparameters in the extended Kalman filter are then input into the improved particle swarm optimization algorithm, and the optimal solution is obtained through multiple iterations.

[0025] Preferably, in step S24, the collected dual-band infrared radiation data is input into the extended Kalman filter with multi-kernel generalized correlation entropy induced function under optimal parameters to obtain the attitude information of the rotating aircraft. The method is as follows:

[0026] Step 241: Based on the roll angle, rotational speed, pitch angle, and time step of the rotating aircraft, set the state matrix of the multi-kernel generalized correlation entropy induced function extended Kalman filter estimation algorithm as follows: The state transition matrix is ​​represented as ;

[0027] Step 242: The multi-kernel generalized correlation entropy induced function extended Kalman filter estimation algorithm includes a prediction step and an update step;

[0028] The prediction step begins with the state equation, based on the state at time k-1. Estimate the state at time k And predict the noise covariance matrix. ;

[0029] The observation matrix is ​​represented as The linearized output matrix is ​​obtained based on the dual-band infrared attitude measurement model. ;

[0030] The update step includes predicting the output matrix. Noise covariance matrix Calculate the Kalman gain K using the multi-kernel generalized correlation entropy induced function;

[0031] Step 243: Adjust the estimated attitude parameters and update the state matrix using the gain matrix K. :

[0032] Step 244: Update the noise covariance matrix of the filter. .

[0033] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0034] 1. The dual-band infrared radiation attitude estimation method of this invention has been verified by semi-physical experiments to demonstrate its superior performance. Compared with existing methods, the mean square error of the roll angle is 0.00008 and the mean square error of the pitch angle is 0.00028.

[0035] 2. The present invention provides a dual-band infrared radiation attitude estimation method, which analyzes the noise characteristics of dual-band infrared radiation data in practical application scenarios and proves that the probability density of the noise is a non-Gaussian distribution.

[0036] 3. The dual-band infrared radiation attitude estimation method of this invention selects the Laplace kernel and Cauchy kernel as the generalized correlation entropy induction function, designs an adaptive kernel size mechanism, and uses an optimized particle swarm optimization algorithm for iteration, thereby achieving high-precision and robust dual-band infrared radiation attitude measurement. Attached Figure Description

[0037] Figure 1 This is a flowchart of the dual-band infrared radiation attitude estimation method of the present invention;

[0038] Figure 2 This is a diagram showing the output data of the infrared radiation sensor in an embodiment of the present invention;

[0039] Figure 3 This is a diagram of dual-band infrared radiation noise data in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram showing the results of 50 iterations of the improved particle swarm optimization algorithm in an embodiment of the present invention;

[0041] Figure 5 This is a schematic diagram showing the error angles of the roll and pitch angles output in an embodiment of the present invention. Detailed Implementation

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings, so that those skilled in the art can better understand the advantages and features of the present invention, thereby making a clearer definition of the scope of protection of the present invention. The embodiments described in this invention are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0043] Meanwhile, the step numbers in the embodiments of the present invention are set only for ease of explanation and do not limit the order between steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0044] In one embodiment of the present invention, a dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering is described, the steps of which are as follows: Figure 1 As shown below, the steps are explained in detail:

[0045] S11. Collect dual-band infrared radiation data under different radiation backgrounds using a hardware-in-the-loop experimental platform and perform data preprocessing. Record the set roll and pitch angle parameters, and use the roll and pitch angle parameters to deduce the theoretical dual-band infrared radiation data. Remove the theoretical data from the actual measurement data to obtain the pure noise data.

[0046] S12. Dual-band infrared radiation attitude noise is caused by factors such as sensor measurement noise, assembly position and motor interference, and has characteristics such as random values, limited amplitude and non-time sequence.

[0047] The impact on dual-band infrared radiation attitude information is that a random noise with a limited amplitude is superimposed on the true attitude value. By using the D'Agostino-Pearson normality test algorithm to judge the pure noise data, the results show that the dual-band infrared radiation attitude noise data is a non-Gaussian or non-zero mean Gaussian distribution, and the noise reduces the reliability of dual-band infrared radiation attitude.

[0048] S21. For estimation algorithms based on maximum correlation entropy, the type of kernel function is crucial.

[0049] Kernel functions map the original error space to the reproducing kernel Hilbert space using Mercer's theorem, making the maximum correlation entropy criterion equivalent to achieving nonlinear separation of signal features in the reproducing kernel Hilbert space, significantly improving robustness against non-Gaussian noise. Single-kernel function types may encounter problems when processing complex data, potentially leading to performance degradation.

[0050] Within the effective data range, the impact of kernel size on filtering performance and convergence speed is contradictory for single-kernel types, which poses a challenge to achieving a balance between performance and speed in a single kernel function.

[0051] In this embodiment, the above problems can be well solved by using a multi-kernel function. The multi-kernel function selected is the Cauchy function and the Laplace kernel, which have good performance in resisting impulse noise and are also less sensitive to parameters.

[0052] S22. The choice of kernel size has a significant impact on the estimation algorithm based on maximum correlation entropy.

[0053] However, engineering practice typically employs fixed-size cores, which are selected through trial and error based on experience for specific types of non-Gaussian noise. Noise is unstable, and initial strong noise tends to plateau over time. Therefore, using a fixed core size can hinder the MCC filter from achieving optimal performance.

[0054] In this embodiment, this problem can be effectively solved by selecting an adaptive kernel size mechanism.

[0055] S23. In the process of selecting the type and size of the kernel function, parameters need to be preset, and the same applies to the extended Kalman filter. These preset parameters play a key role in the algorithm performance, but they are all set based on engineering experience. This not only relies on engineering experience but also requires improvement in accuracy. However, this problem can be well solved through intelligent swarm optimization iteration.

[0056] S24. By inputting the collected dual-band infrared radiation data into the extended Kalman filter of the multi-core generalized correlation entropy induced function under optimal parameters, the attitude information of the rotating aircraft can be obtained, thereby improving the accuracy and robustness of dual-band infrared radiation attitude measurement technology under strong interference.

[0057] Specifically, in step S11, such as Figure 2 As shown, the sampling rate of the hardware-in-the-loop experimental platform was set to 10kHz and the sampling time was 10s to collect infrared radiation data.

[0058] The main preprocessing step for the data is normalization. First, the dual-band infrared radiation data is normalized to the (-1, 1) interval, as shown in the following formula:

[0059] ;

[0060] In the formula, For the input dual-band infrared radiation data, and These are the minimum and maximum values ​​in the input matrix, respectively. This is the output data after normalization.

[0061] Next, the dataset is cleaned by removing unusable data.

[0062] Furthermore, by controlling the attitude changes of the rotating aircraft with motors, roll and pitch angle information can be obtained.

[0063] In this embodiment, the dual-band infrared radiation data is derived by using the following formula:

[0064] ;

[0065] ;

[0066] In the formula, The pitch angle, This is the roll angle. and Two long-wavelength sensors are vertically distributed. and Two medium-wave sensors are vertically distributed.

[0067] Furthermore, actual dual-band data will be collected. And theoretically interference-free data The difference is calculated using the following formula:

[0068] ;

[0069] In the formula, Noise for dual-band infrared radiation measurements.

[0070] Specifically, in step S12, such as Figure 3 As shown, from a time series perspective, noise values ​​exhibit characteristics such as randomness, amplitude limitation, and non-temporal order. The impact on dual-wave infrared radiation attitude information is due to the superposition of a random and... Amplitude-limited noise.

[0071] Furthermore, the D'Agostino-Pearson normality test algorithm is used to assess purely noisy data. The D'Agostino-Pearson normality test algorithm first calculates skewness and kurtosis, using the following formulas:

[0072] ;

[0073] ;

[0074] In the formula, n is the sample size. It is the noise mean. It is the noise standard deviation. and These are skewness and kurtosis, respectively.

[0075] The raw values ​​of skewness and kurtosis depend on the sample size and the dimensions of the data, and using these values ​​directly may make comparisons inconvenient. By transforming them, skewness and kurtosis can be standardized, making them comparable across different datasets and sample sizes.

[0076] The standardization formula is as follows:

[0077] ;

[0078] ;

[0079] In the formula, and These are the standardized skewness and kurtosis, respectively.

[0080] Finally, the D'Agostino statistic is calculated using skewness and kurtosis, as shown in the following formula:

[0081] ;

[0082] In the formula, K is the D'Agostino statistic. The calculation results are checked against the chi-square distribution table, and the results show that the dual-band infrared radiation noise is non-Gaussian distributed.

[0083] Specifically, in step S21, the multi-kernel functions of the extended Kalman filter estimation algorithm for the multi-kernel generalized correlation entropy induced function in this embodiment are selected as the Laplace kernel and the Cauchy kernel, respectively, as shown in the following formula:

[0084] ;

[0085] ;

[0086] In the formula, Represents the core size. It is a natural exponential function. and They are the Laplace nucleus and the Cauchy nucleus, respectively. and These are the heavy-tailed factors of the Laplace nucleus and the Cauchy nucleus, respectively. This represents the input and target values ​​of the function.

[0087] Specifically, in step S22, the extended Kalman filter estimation algorithm of the multi-kernel generalized correlation entropy induced function in this embodiment adds an adaptive kernel size mechanism, which gets rid of the limitation of fixed kernel size.

[0088] The adaptive kernel size mechanism is defined by the following formula:

[0089] ;

[0090] In the formula, This is the inverse transformation of the observation noise covariance matrix.

[0091] Specifically, in step S23, such as Figure 4 As shown, the improved particle swarm optimization algorithm is selected as the intelligent swarm optimization algorithm for iterative updates.

[0092] In this embodiment, the improvements of the improved particle swarm optimization algorithm compared to the traditional particle swarm optimization algorithm mainly consist of two parts:

[0093] 1) Introduce a linear differential decreasing strategy for inertia weight. This is to adjust the inertia weight at different search stages to avoid getting trapped in local optima.

[0094] The complete 50-iteration process is as follows: First, a linear differential decreasing strategy for the inertia weight is set, as shown in the following formula:

[0095] ;

[0096] In the formula, For inertial weights, For maximum inertia weight, The minimum inertia weight; This represents the current iteration number. is the maximum number of iterations; n is a power parameter that controls the non-linearity of the weight changes.

[0097] Next, the particle velocity v is updated using the following formula:

[0098] ;

[0099] In the formula, This represents the best historical position of an individual particle. The optimal position for the entire population; and As a learning factor, , ∈[0,1] is a random number.

[0100] 2) In order to enhance the ability of particles to escape local optima, a simulated annealing strategy was introduced into the particle swarm optimization algorithm.

[0101] The formula for updating the particle's position x is as follows:

[0102] ;

[0103] In the formula, For particle position, This represents the particle velocity.

[0104] After the position is updated, the Metropolis criterion of simulated annealing is used to determine whether to accept the new solution. The criterion is shown in the following formula:

[0105] ;

[0106] In the formula, The degree to which new values ​​are received; The annealing temperature is the current temperature, which decreases linearly with each iteration. The initial temperature; This represents the fitness difference between the new solution and the current solution.

[0107] In this embodiment, .

[0108] Furthermore, by inputting the hyperparameters of the extended Kalman filter with the multi-kernel generalized correlation entropy induced function into the improved particle swarm optimization algorithm, the optimal solution can be obtained, and all subsequent parameters are optimal parameter results.

[0109] Specifically, in step S24, such as Figure 5 As shown, the multi-kernel generalized correlation entropy induced function extended Kalman filter estimation (MICI_EKF) algorithm under optimal parameters is described below:

[0110] First, the state matrix of the extended Kalman filter estimation algorithm with multi-kernel generalized correlation entropy induced function is set as follows: ,in, This is the roll angle. For rotational speed, The pitch angle.

[0111] The state transition matrix is ​​set as follows: ,in, For time step.

[0112] Furthermore, the multi-kernel generalized correlation entropy induced function extended Kalman filter estimation algorithm is an improved algorithm on EKF, and it still includes two parts: a prediction step and an update step.

[0113] The prediction step starts with the state equation and estimates the state at time k based on the state at time k-1.

[0114] .

[0115] The prediction noise covariance matrix is ​​obtained by the following formula:

[0116] ;

[0117] In the formula, To predict the noise covariance matrix, Let be the system noise covariance matrix.

[0118] The observation matrix is ​​represented as Y:

[0119] ;

[0120] in, , These are data from long-wave infrared x-axis and y-axis sensors, respectively. , These are mid-wave infrared sensor data for the x and y axes, respectively.

[0121] The linearized output matrix is ​​obtained based on the dual-band infrared attitude measurement model. for:

[0122] .

[0123] The update steps include the predicted output matrix H and the noise covariance matrix. Calculate Kalman gain using the multi-kernel generalized correlation entropy induced function. The formula is as follows:

[0124] ;

[0125] In the formula, Represents the observation noise variance matrix. The MCIC coefficient is calculated using the following formula:

[0126] ;

[0127] In the formula, The weights are Laplace kernel weights.

[0128] Gain matrix Adjust the estimated attitude parameters and update the state matrix. The formula is as follows:

[0129] ;

[0130] In the formula, Indicates the sensor's time Actual measurement data at the location, Indicates the sensor's time Estimated data at the location.

[0131] Finally, the noise covariance matrix of the filter Updated by the following formula:

[0132] ;

[0133] In the formula, Indicates the sensor's time Kalman gain data at the location.

[0134] In summary, this invention presents a dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering, and its superior performance has been verified through semi-physical experiments. Figure 4 As can be seen, the mean square error of the roll angle in the steady state is calculated to be 0.00008 and the mean square error of the pitch angle is 0.00028 by the method of the present invention, realizing high-precision and robust dual-band infrared radiation attitude measurement.

[0135] The above description is merely a preferred embodiment of the present invention. It should be noted that the descriptions and practices disclosed in this invention are readily conceived and understood by those skilled in the art, and various improvements and modifications can be made without departing from the principles of the present invention. Therefore, modifications or improvements made without departing from the spirit of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A dual-band infrared radiation attitude estimation method using multi-kernel correlation entropy extended Kalman filtering, characterized in that, Includes the following steps: S11. Collect dual-band infrared radiation data under different radiation backgrounds through a semi-physical experimental platform and perform data preprocessing; by recording the roll angle and pitch angle parameter information set by the rotating aircraft, deduce the theoretical dual-band infrared radiation data, remove the theoretical data from the actual collected data, and obtain pure noise data. S12. Analyze the probability distribution characteristics of the actual measured dual-band infrared radiation noise, and use the D'Agostino-Pearson normality test algorithm to judge the pure noise data. From a time series perspective, the pure noise data has the characteristics of numerical randomness, amplitude limitation, and non-temporal order. Its impact on the dual-wave infrared radiation attitude information is that it superimposes a random and... Amplitude-limited noise; using the D'Agostino-Pearson normality test algorithm to determine the pure noise data, including the following sub-steps: Step S121: Calculate the skewness and kurtosis of the noise: ； ； In the formula, It's the sample size. It is the noise mean. It is the noise standard deviation. and These are skewness and kurtosis, respectively. Step S122: Standardize the skewness and kurtosis: ； ； In the formula, and These are the standardized skewness and kurtosis, respectively. Step S123: Calculate the D'Agostino statistic using skewness and kurtosis: ； In the formula, K is the D'Agostino statistic. The calculation results are checked against the chi-square distribution table, and the results show that the dual-band infrared radiation noise is non-Gaussian distributed. S21. Construct a multi-kernel generalized correlation entropy induced function extended Kalman filter estimation algorithm, and select the Laplace kernel and Cauchy kernel as the kernel functions of the multi-kernel correlation entropy extended Kalman filter; The Laplace nucleus and Cauchy nucleus are defined by the following formulas: ； ； In the formula, Represents the core size. It is a natural exponential function. and They are the Laplace nucleus and the Cauchy nucleus, respectively. and These are the heavy-tailed factors of the Laplace nucleus and the Cauchy nucleus, respectively. This represents the input and target values ​​of the function; S22, The multi-core correlation entropy extended Kalman filter selects an adaptive kernel size mechanism to get rid of the fixed kernel size limitation; The adaptive kernel size mechanism is defined by the following formula: ； In the formula, Inverse transformation of the observation noise covariance matrix; S23. The optimal parameters of the multi-core correlated entropy extended Kalman filter are replaced by improving the particle swarm optimization algorithm. The linear differential decreasing strategy of inertial weight is introduced to adjust the inertial weight in different search stages. Combined with the simulated annealing strategy, the ability of particles to jump out of local optima is enhanced. S24. Input the collected dual-band infrared radiation data into the extended Kalman filter with the multi-core generalized correlation entropy induced function under the optimal parameters to obtain the attitude information of the rotating aircraft.

2. The dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering according to claim 1, characterized in that, In step S11, the sampling rate and sampling time interval of the hardware-in-the-loop experimental platform are set, dual-band infrared radiation data under different radiation backgrounds are collected, the collected data are normalized, the dual-band infrared radiation data are unified into the (-1,1) interval, the data is cleaned, and unusable data is deleted. The normalization process is described in the following formula: ； In the formula, For the input dual-band infrared radiation data, and These are the minimum and maximum values ​​in the input matrix, respectively. This is the normalized output of dual-band infrared radiation data.

3. The dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering according to claim 2, characterized in that, In step S11, the attitude change of the rotating aircraft is controlled by the motor to obtain roll angle and pitch angle information, and the dual-band infrared radiation data is derived from it, as shown in the following formula: ； ； In the formula, The pitch angle, This is the roll angle. and Two long-wavelength sensors are vertically distributed. and Two medium-wave sensors are vertically distributed.

4. The dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering according to claim 3, characterized in that, In step S11, pure noise data is obtained, and the actual collected dual-band infrared radiation data is... And theoretically interference-free data The difference is calculated using the following formula: ； In the formula, This is the measured pure noise data for dual-band infrared radiation.

5. The dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering according to claim 1, characterized in that, In step S23, a linear differential decreasing strategy for inertia weights is introduced to adjust the inertia weights at different search stages, including the following sub-steps: Step 231: Set the linear differential decreasing strategy for inertia weights, as shown in the following formula: ； In the formula, For inertial weights, For maximum inertia weight, The minimum inertia weight; This represents the current iteration number. is the maximum number of iterations; n is a power parameter that controls the non-linearity of the weight changes; Step 232: Update the particle velocity v. For the first... The formula for the next iteration is as follows: ； In the formula, This represents the optimal position in the history of an individual particle. The optimal position for the global population. and As a learning factor, , ∈[0,1] represents a random number; Step 233, the formula for updating the particle's position x is as follows: ； In the formula, For particle position, This represents the particle velocity.

6. The dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering according to claim 5, characterized in that, In step S23, the ability of particles to escape local optima is enhanced by combining simulated annealing strategy, as follows: After the particle's position is updated, the Metropolis criterion of simulated annealing is used to determine whether to accept the new solution: ； In the formula, To what extent new values ​​are received, The annealing temperature is the current temperature, which decreases linearly with each iteration. This represents the fitness difference between the new solution and the current solution. The hyperparameters in the extended Kalman filter are obtained by extending the multi-kernel generalized correlation entropy induced function, and then inputting them into the improved particle swarm optimization algorithm. Through multiple iterations, the optimal solution is obtained.

7. The dual-band infrared radiation attitude estimation method based on multi-kernel correlation entropy extended Kalman filtering according to claim 6, characterized in that, In step S24, the collected dual-band infrared radiation data is input into the extended Kalman filter with multi-kernel generalized correlation entropy induced function under optimal parameters to obtain the attitude information of the rotating aircraft. The method is as follows: Step 241: Set the state matrix of the extended Kalman filter estimation algorithm with the multi-kernel generalized correlation entropy induced function as... The state transition matrix is ​​represented as : ； ； in, This is the roll angle. For rotational speed, The pitch angle, For time step, superscript T Indicates transpose; Step 242: The multi-kernel generalized correlation entropy induced function extended Kalman filter estimation algorithm includes a prediction step and an update step; The prediction step begins with the state equation, based on... State estimation at time 1 The state at any given moment: ； Prediction noise covariance matrix : ； In the formula, The noise covariance matrix is... The system noise covariance matrix; The observation matrix is ​​represented as : ； in, , These are data from long-wave infrared x-axis and y-axis sensors, respectively. , These are mid-wave infrared sensor data for the x and y axes, respectively. The linearized output matrix is ​​obtained based on the dual-band infrared attitude measurement model. : ； The update step includes predicting the output matrix. Noise covariance matrix Calculate the Kalman gain using the multi-kernel generalized correlation entropy induced function. : ； In the formula, Represents the observation noise variance matrix. The MCIC coefficient; Step 243: Using the gain matrix Adjust the estimated attitude parameters and update the state matrix. : ； Step 244: Update the noise covariance matrix of the filter. : ； In the formula, Indicates that the sensor is in Actual measurement data at time [time] Indicates that the sensor is in Estimated data at time t, Indicates that the sensor is in Kalman gain data at time point 1.

Images