A multi-source photoelectric detection data fusion method, device, equipment and medium

Abstract

The present invention relates to a multi-source photoelectric detection data fusion method, device, equipment and medium, belonging to the field of target detection data fusion technology, and solves the problem that the data of multi-source photoelectric detection equipment has low fusion data accuracy due to non-Gaussian noise. The technical solution of the present invention mainly includes: obtaining the posterior distribution of the state quantity of the target through Bayesian inference based on the state transition model and the observation model, wherein the state transition model includes non-Gaussian state noise, and the observation model includes non-Gaussian observation noise; setting the number of Gaussian components of the Gaussian mixture model to the number of photoelectric detection devices in the multi-source photoelectric detection system, and expanding and approximating the posterior distribution through the Gaussian mixture model to determine the mean and variance of each Gaussian component; using the mean of each Gaussian component as an unbiased estimate, and performing data fusion optimization through a subgradient fusion algorithm to obtain a fused estimate of the target state quantity.

Description

Technical Field

[0001] The present invention belongs to the technical field of target detection data fusion, and specifically relates to a multi-source photoelectric detection data fusion method, device, equipment and medium. Background Art

[0002] Photoelectric detection equipment has the advantages of high power, small size, and easy portability. It can quickly complete tasks such as target detection, detection, and tracking. With the rapid development and in-depth application of drone swarm technology, airborne photoelectric detection equipment has played an important role in the military and civilian fields. However, unlike vehicle-mounted or handheld photoelectric detection equipment, airborne photoelectric detection equipment will move at high speed with the drone. In the application scenario of drone swarms, the detection distance, detection angle, detection altitude, atmosphere, lighting, and interference faced by each airborne photoelectric detection equipment are different. There may even be photoelectric detection equipment with different models, precision, and applicable conditions in the entire drone swarm. Overall, the noise of the multi-source detection data obtained from the entire drone swarm shows strong coupling, nonlinearity, fast time-varying, and non-Gaussian characteristics, and traditional data fusion algorithms cannot be used directly.

[0003] In summary, the existing technology has a problem of low accuracy of fused data caused by non-Gaussian noise for data from multi-source photoelectric detection equipment, which needs to be solved. Summary of the invention

[0004] In view of the above analysis, the embodiments of the present invention aim to provide a multi-source photoelectric detection data fusion method, device, equipment and medium to solve the problem of low fusion data accuracy caused by non-Gaussian noise in the data of multi-source photoelectric detection equipment.

[0005] The embodiment of the present invention provides a multi-source photoelectric detection data fusion method, comprising the following steps:

[0006] S1. Obtaining the posterior distribution of the state quantity of the target through Bayesian inference based on the state transfer model and the observation model, wherein the state transfer model includes non-Gaussian state noise, and the observation model includes non-Gaussian observation noise.

[0007] S2, setting the number of Gaussian components of the Gaussian mixture model to the number of photoelectric detection devices in the multi-source photoelectric detection system, and performing an expansion approximation on the posterior distribution through the Gaussian mixture model to determine the mean and variance of each Gaussian component;

[0008] S3. Taking the mean of each Gaussian component as an unbiased estimate, performing data fusion optimization through a subgradient fusion algorithm to obtain a fused estimate of the target state quantity.

[0009] In some embodiments, in step S1, obtaining the posterior distribution of the state quantity of the target by Bayesian inference based on the state transition model and the observation model includes:

[0010] S11. Establish the state transition model and observation model of the system, expressed as:

[0011] , where k represents the sampling point, represents the target state at sampling point k, represents the state transition model at sampling point k+1, represents non-Gaussian state noise, represents the system observation of the target at sampling point k, represents the observation model at sampling point k, represents non-Gaussian observation noise;

[0012] S12, obtaining the posterior distribution of the state quantity according to Bayesian inference based on the state transfer model and the observation model;

[0013] S13, performing weighted approximation on the non-Gaussian state noise and the non-Gaussian observation noise through a Gaussian probability density function group to obtain approximate representations of the non-Gaussian state noise and the non-Gaussian observation noise respectively;

[0014] S14. Obtain an approximate representation of the posterior distribution based on the approximate representation of each non-Gaussian noise and the relationship between each non-Gaussian noise and the posterior distribution.

[0015] In some embodiments, in step S12, obtaining the posterior distribution of the state quantity according to Bayesian inference based on the state transition model and the observation model includes:

[0016] Based on the posterior distribution of the previous sampling point and the state transition model, the prior distribution of the current sampling point is obtained by prediction, which is expressed as:

[0017] , Represents the observation based on the last sampling point The prior distribution of represents the state quantity distribution at the sampling point k obtained by the state transfer model, Represents the posterior distribution of the state quantity at the previous sampling point k-1;

[0018] The prior distribution of the current sampling point is updated based on the observation model to obtain the posterior distribution of the current sampling point, which is expressed as:

[0019] ,in Indicates that the sampling point k is based on the target state quantity through the observation model The observed value distribution of Represents the posterior distribution at the current sampling point k.

[0020] In some embodiments, in step S13, weighted approximation is performed on the non-Gaussian state noise and the non-Gaussian observation noise respectively by using a Gaussian probability density function group to obtain an approximate representation of each of the non-Gaussian state noise and the non-Gaussian observation noise, including:

[0021] Construct a group of Gaussian probability density functions with m groups of means and variances, expressed as:

[0022] , represents the group of Gaussian probability density functions, where represents the mean vector of the i-th component in the Gaussian probability density function group, represents the covariance matrix of the i-th component in the Gaussian probability density function group, and n represents the dimension of the vector x;

[0023] Performing weighted summation on the components of the Gaussian probability density function group to obtain respective approximations of the non-Gaussian state noise and the non-Gaussian observation noise;

[0024] The approximate expression of the non-Gaussian state noise is:

[0025] ,in Non-Gaussian state noise An approximate representation of For approximation The number of Gaussian probability density functions, is the weight corresponding to the i-th component in the approximate representation of non-Gaussian state noise;

[0026] The non-Gaussian observation noise is approximately expressed as:

[0027] ,in Non-Gaussian state noise An approximate representation of For approximation The number of Gaussian probability density functions, Represents the weight corresponding to the i-th component in the approximate representation of non-Gaussian observation noise.

[0028] In some embodiments, in step S14, obtaining an approximate representation of the posterior distribution based on the approximate representation of each non-Gaussian noise and the relationship between each non-Gaussian noise and the posterior distribution includes:

[0029] The posterior distribution of the previous sampling point is approximated by a Gaussian probability density function group, which is expressed as:

[0030] ,in represents the posterior distribution of the previous sampling point k-1, is the weight corresponding to the i-th component in the approximate representation of the posterior distribution of the previous sampling point;

[0031] Substituting the approximate representation of non-Gaussian state noise into the state transfer model to approximate the state transfer model is expressed as:

[0032] ;

[0033] Substituting the approximate representation of non-Gaussian observation noise into the observation model to obtain an approximation of the distribution of observations obtained by the observation model is expressed as:

[0034] ;

[0035] Substitute the approximation of the state transition model and the posterior distribution of the previous sampling point k-1 into the prior distribution The approximation of the prior distribution is obtained as:

[0036] ;

[0037] The approximation of the prior distribution and the approximation of the observed value distribution are substituted into the posterior distribution to obtain an approximate representation of the posterior distribution.

[0038] In some embodiments, the method of using the mean of each Gaussian component as an unbiased estimate and performing data fusion optimization through a subgradient fusion algorithm to obtain a fused estimate of the target state quantity includes:

[0039] The fused estimated value is set to be the sum of the products of the fusion weight matrices corresponding to each unbiased estimated value and the unbiased estimated value, and the sum of the fusion weight matrices is satisfied to be a unit matrix;

[0040] The optimization objective function is determined to minimize the mean square error of the fused estimate, and the initial mean square error is obtained based on the covariance matrix or cross-covariance matrix between the initial fusion weight matrix and each unbiased estimate:

[0041] Calculating the sub-gradient of the objective function for each of the fusion weight matrices;

[0042] Updating the fusion weight matrix based on the subgradient;

[0043] Calculate a new mean square error based on the updated fusion weight matrix. If the new mean square error is less than the mean square error before this iteration, update the mean square error to the mean square error obtained in this iteration. Otherwise, retain the mean square error and fusion weight matrix before the update.

[0044] Iteratively updating the fusion weight matrix until a preset condition is met to obtain a final fusion weight matrix;

[0045] A final fusion estimate of the target state quantity is obtained based on the final fusion weight matrix and the corresponding unbiased estimate.

[0046] In some embodiments, the objective function is equivalently determined as:

[0047] ,

[0048] in represents the trace of the matrix, represents the fusion weight matrix of the i-th unbiased estimate, represents the covariance matrix of the i-th unbiased estimate, represents the cross-covariance matrix between the ith unbiased estimator and the jth unbiased estimator.

[0049] The present invention also provides a multi-source photoelectric detection data fusion device, comprising:

[0050] An inference module, which obtains a posterior distribution of a state quantity of a target through Bayesian inference based on the state transition model and the observation model, wherein the state transition model includes non-Gaussian state noise and the observation model includes non-Gaussian observation noise;

[0051] An approximation module, which sets the number of Gaussian components of the Gaussian mixture model to the number of photoelectric detection devices in the multi-source photoelectric detection system, and approximates the posterior distribution by expanding the Gaussian mixture model to determine the mean and variance of each Gaussian component;

[0052] The fusion module takes the mean of each Gaussian component as an unbiased estimate, performs data fusion optimization through a subgradient fusion algorithm, and obtains a fused estimate of the target state quantity.

[0053] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the multi-source photoelectric detection data fusion method as described in any of the above embodiments is implemented.

[0054] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the multi-source photoelectric detection data fusion method as described in any of the above embodiments is implemented.

[0055] The embodiments of the present invention have at least the following beneficial effects:

[0056] The present invention firstly approximates the state quantity of the system according to the state transfer model and the observation model through Bayesian inference, specifically involves approximating the non-Gaussian state noise and non-Gaussian observation noise contained therein and then calculating them to obtain the approximation of the posterior distribution of the state quantity, and then expands the approximate fitting of the posterior distribution into several Gaussian components through the Gaussian mixture model. Therefore, the posterior distribution can be considered as the superposition of unbiased observations of multiple photoelectric detection devices, that is, to obtain several unbiased estimates, and then the obtained unbiased estimates are optimized and fused through the subgradient fusion algorithm to obtain a fused estimate. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this specification. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0058] Figure 1 A schematic diagram of the multi-source photoelectric detection data fusion method provided by the present invention;

[0059] Figure 2 A schematic diagram of a Gaussian mixing process involved in an embodiment of the present invention;

[0060] Figure 3 A schematic diagram of a multi-source photoelectric detection data fusion device provided in an embodiment of the present invention;

[0061] Figure 4 A schematic diagram of an electronic device architecture provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. It should be noted that, in the absence of conflict, the embodiments and features in the embodiments disclosed in this disclosure can be combined, separated, interchanged and / or rearranged with each other. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0063] The terms used here are for the purpose of describing specific embodiments, and are not intended to be restrictive. As used here, unless the context clearly indicates otherwise, the singular forms "one (kind, person)" and "said (the)" are also intended to include plural forms. In addition, when the terms "comprise" and / or "include" and their variations are used in this specification, it is explained that there are stated features, integral bodies, steps, operations, parts, assemblies and / or their groups, but it is not excluded that there are or add one or more other features, integral bodies, steps, operations, parts, assemblies and / or their groups. It should also be noted that, as used here, the terms "substantially", "approximately" and other similar terms are used as approximate terms and not as degree terms, so that they are used to explain the inherent deviations of the measured values, calculated values ​​and / or the values ​​provided that will be recognized by those of ordinary skill in the art.

[0064] The embodiments of the present invention are intended to provide a multi-source photoelectric detection data fusion method, device, equipment and medium to solve the problem of low fusion data accuracy due to non-Gaussian noise in the data of multi-source photoelectric detection equipment. The application scenario of the present invention is, for example, the data optimization problem of multi-source photoelectric detection equipment based on unmanned clusters. The embodiments of the present invention are also a data fusion method for multi-source photoelectric detection equipment of unmanned aerial vehicle swarms.

[0065] The embodiment of the present invention first provides a multi-source photoelectric detection data fusion method, such as Figure 1 As shown, the following steps are included:

[0066] S1. Obtaining the posterior distribution of the state quantity of the target through Bayesian inference based on the state transfer model and the observation model, wherein the state transfer model includes non-Gaussian state noise, and the observation model includes non-Gaussian observation noise.

[0067] S2. The number of Gaussian components of the Gaussian mixture model is set to the number of photoelectric detection devices in the multi-source photoelectric detection system, and the posterior distribution is approximated by expanding the Gaussian mixture model to determine the mean and variance of each Gaussian component.

[0068] S3. Taking the mean of each Gaussian component as an unbiased estimate, performing data fusion optimization through a subgradient fusion algorithm to obtain a fused estimate of the target state quantity.

[0069] It should be understood that the embodiment of the present invention involves processing in a time step in the entire observation process. Each time step will obtain the state quantity and noise measured by the multi-source photoelectric detection data. In addition, at different time steps, due to changes in the target state, the state transfer model and the observation model itself may also change. The present invention optimizes the data obtained by the multi-source photoelectric detection equipment through the above method to obtain a more accurate target state quantity.

[0070] In some embodiments, in step S1, obtaining the posterior distribution of the state quantity of the target by Bayesian inference based on the state transition model and the observation model includes:

[0071] S11. Establish the state transition model and observation model of the system, expressed as:

[0072] , where k represents the sampling point, represents the target state at sampling point k, represents the state transition model at sampling point k+1, represents non-Gaussian state noise, represents the system observation of the target at sampling point k, represents the observation model at sampling point k, represents non-Gaussian observation noise. The sampling point k is also the kth time step.

[0073] S12. Obtaining the posterior distribution of the state quantity according to Bayesian inference based on the state transfer model and the observation model.

[0074] In some embodiments, in step S12, obtaining the posterior distribution of the state quantity according to Bayesian inference based on the state transition model and the observation model includes:

[0075] Based on the posterior distribution of the previous sampling point and the state transition model, the prior distribution of the current sampling point is obtained by prediction, which is expressed as:

[0076] , Represents the observation based on the last sampling point The prior distribution of represents the state quantity distribution at the sampling point k obtained by the state transfer model, Represents the posterior distribution of the state quantity at the previous sampling point k-1;

[0077] The prior distribution of the current sampling point is updated based on the observation model to obtain the posterior distribution of the current sampling point, which is expressed as:

[0078] ,in Indicates that the sampling point k is based on the target state quantity through the observation model The observed value distribution of Represents the posterior distribution at the current sampling point k.

[0079] S13, weighted approximation is performed on the non-Gaussian state noise and the non-Gaussian observation noise by using a Gaussian probability density function group to obtain approximate representations of the non-Gaussian state noise and the non-Gaussian observation noise. Preferably, in some embodiments, only approximation can be performed by using a Gaussian mixture model.

[0080] In some embodiments, in step S13, weighted approximation is performed on the non-Gaussian state noise and the non-Gaussian observation noise respectively by using a Gaussian probability density function group to obtain an approximate representation of each of the non-Gaussian state noise and the non-Gaussian observation noise, including:

[0081] Construct a group of Gaussian probability density functions with m groups of means and variances, expressed as:

[0082] , represents the group of Gaussian probability density functions, where represents the mean vector of the i-th component in the Gaussian probability density function group, represents the covariance matrix of the i-th component in the Gaussian probability density function group, and n represents the dimension of the vector x;

[0083] Performing weighted summation on the components of the Gaussian probability density function group to obtain respective approximations of the non-Gaussian state noise and the non-Gaussian observation noise;

[0084] The approximate expression of the non-Gaussian state noise is:

[0085] ,in Non-Gaussian state noise An approximate representation of For approximation The number of Gaussian probability density functions, is the weight corresponding to the i-th component in the approximate representation of non-Gaussian state noise;

[0086] The non-Gaussian observation noise is approximately expressed as:

[0087] ,in Non-Gaussian state noise An approximate representation of For approximation The number of Gaussian probability density functions, Represents the weight corresponding to the i-th component in the approximate representation of non-Gaussian observation noise.

[0088] S14. Obtain an approximate representation of the posterior distribution based on the approximate representation of each non-Gaussian noise and the relationship between each non-Gaussian noise and the posterior distribution.

[0089] In some embodiments, in step S14, obtaining an approximate representation of the posterior distribution based on the approximate representation of each non-Gaussian noise and the relationship between each non-Gaussian noise and the posterior distribution includes:

[0090] The posterior distribution of the previous sampling point is approximated by a Gaussian probability density function group, which is expressed as:

[0091] ,in represents the posterior distribution of the previous sampling point k-1, is the weight corresponding to the i-th component in the approximate representation of the posterior distribution of the previous sampling point;

[0092] Substituting the approximate representation of non-Gaussian state noise into the state transfer model to approximate the state transfer model is expressed as:

[0093] ;

[0094] Substituting the approximate representation of non-Gaussian observation noise into the observation model to obtain an approximation of the distribution of observations obtained by the observation model is expressed as:

[0095] ;

[0096] Substitute the approximation of the state transition model and the posterior distribution of the previous sampling point k-1 into the prior distribution The approximation of the prior distribution is obtained as:

[0097] ;

[0098] The approximation of the prior distribution and the approximation of the observed value distribution are substituted into the posterior distribution to obtain an approximate representation of the posterior distribution.

[0099] It should be understood that the process of using the Gaussian mixture model to approximate the known noise or posterior distribution involved in the present invention is also an iterative process, such as Figure 2 As shown, a partial process of performing Gaussian mixture approximate expansion on the approximate representation of the posterior distribution of the detection data of the system is shown, for example, when the number of photoelectric detection devices in the system is 3.

[0100] In some embodiments, the method of using the mean of each Gaussian component as an unbiased estimate and performing data fusion optimization through a subgradient fusion algorithm to obtain a fused estimate of the target state quantity includes:

[0101] S31. Set the fused estimate to be the sum of the products of the fusion weight matrices corresponding to each unbiased estimate and the unbiased estimate, and satisfy that the sum of the fusion weight matrices is a unit matrix.

[0102] First of all, it should be understood that the unbiased estimate here is often a vector with multi-dimensional values, such as the size, distance, speed and other parameters of the target. In addition, the unbiased estimate is based on the Gaussian component separated by the Gaussian mixture expansion of the posterior distribution in the previous step. Therefore, its mean, variance, covariance of each estimate and the mutual covariance between estimates are all determined. The unbiased estimate can be understood as a noise-free and unbiased measurement value measured by multiple devices, which is optimized and fused by the subgradient optimization fusion algorithm.

[0103] Specifically, assuming that the number of unbiased estimates is N, the dimension is M, and x1, x2..., xN are multiple unbiased estimates of the random quantity target state x, the fused estimate is expressed as:

[0104]

[0105] in, represents the fusion estimate, Ai (i=1, 2,…, N) is the M-dimensional fusion weight matrix, and I is the unit matrix.

[0106] S32. Determine the optimization objective function to minimize the mean square error of the fused estimate, and obtain the initial mean square error based on the covariance matrix or cross-covariance matrix between the initial fusion weight matrix and each unbiased estimate.

[0107] Remember V ii is the covariance matrix of the i-th unbiased estimate, V ij is the mutual covariance matrix of the i-th and j-th unbiased estimates, then the joint covariance matrix of the unbiased estimates is expressed as:

[0108]

[0109] Then, based on this, in some embodiments, the objective function is equivalently determined to be expressed as:

[0110] ,

[0111] in represents the trace of the matrix, represents the fusion weight matrix of the i-th unbiased estimate, represents the covariance matrix of the i-th unbiased estimate, represents the cross-covariance matrix between the i-th unbiased estimator and the j-th unbiased estimator. The objective function is the sum of the matrix traces, which expresses the sum of the weighted variances and correlations (including autocorrelations and cross-correlations). It can be understood as a weight matrix and , integrating covariance information from multiple data sources.

[0112] S33, calculating the sub-gradient of the objective function for each of the fusion weight matrices.

[0113] Preferably, in some embodiments, in order to simplify the calculation of matrix subgradients, the fusion weight matrix can be linearly transformed so that the trace matrix (correlation sum matrix, such as ) can be converted to In the form of, the influence of the cross-covariance matrix is ​​absorbed by the fusion weight matrix.

[0114] Since for any M-dimensional square matrix A, its singular values ​​can be decomposed into A = UΛV T ,in, Λ represents a diagonal matrix, and U and V represent orthogonal matrices. Therefore, A′ j T A′ i Perform a singular decomposition:

[0115]

[0116] The above transformation is expressed as:

[0117] ;

[0118] So define ,in represents the kth singular value of the matrix, then T ji With respect to A′ i and A′ j The subgradient of:

[0119]

[0120] Then define The subgradient of is:

[0121] ;

[0122] It should be understood that That is equivalent to the above objective function

[0123] Depend on( A ' j T A ' i ) T = A ' i T A ' j AvailableC ji = C ij and D ji = D ij ,therefore:

[0124] This is a simplified way to find the subgradient of the objective function.

[0125] In addition, B i = Λ i -0.5 U i T , then .exist Under the constraints, define A′ i ( i =1,2,…, N ) is the projection of P C ( A′ i ):

[0126] .

[0127] S34. Update the fusion weight matrix based on the sub-gradient.

[0128] Specifically, we can first i and A′ j The transformed matrix is ​​updated, and then the updated fusion weight matrix A is calculated i Etc., for example including:

[0129] Select the strategy calculation step length h, let ; If the calculated new mean square error is smaller, the fusion weight matrix is ​​updated, and

[0130] Then S35, a new mean square error is calculated based on the updated fusion weight matrix. If the new mean square error is smaller than the mean square error before this iteration, the mean square error is updated to the mean square error obtained in this iteration. Otherwise, the mean square error and the fusion weight matrix before the update are retained.

[0131] S36, iteratively updating the fusion weight matrix until a preset condition is met to obtain a final fusion weight matrix;

[0132] S37. Obtain a final fusion estimate of the target state quantity based on the final fusion weight matrix and the corresponding unbiased estimate.

[0133] The present invention firstly approximates the state quantity of the system according to the state transfer model and the observation model through Bayesian inference, specifically involves approximating the non-Gaussian state noise and non-Gaussian observation noise contained therein and then calculating them to obtain the approximation of the posterior distribution of the state quantity, and then expands the approximate fitting of the posterior distribution into several Gaussian components through the Gaussian mixture model. Therefore, the posterior distribution can be considered as the superposition of unbiased observations of multiple photoelectric detection devices, that is, to obtain several unbiased estimates, and then the obtained unbiased estimates are optimized and fused through the subgradient fusion algorithm to obtain a fused estimate.

[0134] The present invention also provides a multi-source photoelectric detection data fusion device, such as Figure 3 As shown, including:

[0135] An inference module, which obtains a posterior distribution of a state quantity of a target through Bayesian inference based on the state transition model and the observation model, wherein the state transition model includes non-Gaussian state noise and the observation model includes non-Gaussian observation noise;

[0136] An approximation module, which sets the number of Gaussian components of the Gaussian mixture model to the number of photoelectric detection devices in the multi-source photoelectric detection system, and approximates the posterior distribution by expanding the Gaussian mixture model to determine the mean and variance of each Gaussian component;

[0137] The fusion module takes the mean of each Gaussian component as an unbiased estimate, performs data fusion optimization through a subgradient fusion algorithm, and obtains a fused estimate of the target state quantity.

[0138] The present invention also provides an electronic device, such as Figure 4 As shown, it includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the multi-source photoelectric detection data fusion method as described in any of the above embodiments is implemented.

[0139] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the multi-source photoelectric detection data fusion method as described in any of the above embodiments is implemented.

[0140] Computer readable storage media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0141] Professionals should also be further aware that the operation of the module can be realized by the protocol or program of the prior art, without relying on the new computer program itself. The units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0142] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0143] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-source photoelectric detection data fusion method, characterized in that: include: S1. Obtaining the posterior distribution of the state quantity of the target by Bayesian inference based on the state transfer model and the observation model, wherein the state transfer model includes non-Gaussian state noise, and the observation model includes non-Gaussian observation noise; S2, setting the number of Gaussian components of the Gaussian mixture model to the number of photoelectric detection devices in the multi-source photoelectric detection system, and approximating the posterior distribution by expanding the Gaussian mixture model to determine the mean and variance of each Gaussian component; S3, taking the mean of each Gaussian component as an unbiased estimate, performing data fusion optimization through a subgradient fusion algorithm to obtain a fused estimate of the target state quantity; In step S1, the posterior distribution of the state quantity of the target is obtained by Bayesian inference based on the state transition model and the observation model, including: S11. Establish the state transition model and observation model of the system, expressed as: , where k represents the sampling point, represents the target state at sampling point k, represents the state transition model at sampling point k+1, represents non-Gaussian state noise, represents the system observation of the target at sampling point k, represents the observation model at sampling point k, represents non-Gaussian observation noise, where sampling point k is also the kth time step; S12, obtaining the posterior distribution of the state quantity according to Bayesian inference based on the state transfer model and the observation model; In step S12, obtaining the posterior distribution of the state quantity according to Bayesian inference based on the state transition model and the observation model includes: Based on the posterior distribution of the previous sampling point and the state transition model, the prior distribution of the current sampling point is obtained by prediction, which is expressed as: , Represents the observation based on the last sampling point The prior distribution of represents the state quantity distribution at the sampling point k obtained by the state transfer model, Represents the posterior distribution of the state quantity at the previous sampling point k-1; The prior distribution of the current sampling point is updated based on the observation model to obtain the posterior distribution of the current sampling point, which is expressed as: ,in Indicates that the sampling point k is based on the target state quantity through the observation model The observed value distribution of Represents the posterior distribution at the current sampling point k; S13, performing weighted approximation on the non-Gaussian state noise and the non-Gaussian observation noise through a Gaussian probability density function group to obtain approximate representations of the non-Gaussian state noise and the non-Gaussian observation noise respectively, and only approximating through a Gaussian mixture model; In step S13, weighted approximation is performed on the non-Gaussian state noise and the non-Gaussian observation noise respectively through a Gaussian probability density function group to obtain approximate representations of the non-Gaussian state noise and the non-Gaussian observation noise, including: Construct a group of Gaussian probability density functions with m groups of means and variances, expressed as: , represents the group of Gaussian probability density functions, where represents the mean vector of the i-th component in the Gaussian probability density function group, represents the covariance matrix of the i-th component in the Gaussian probability density function group, and n represents the dimension of the vector x; Performing weighted summation on the components of the Gaussian probability density function group to obtain respective approximations of the non-Gaussian state noise and the non-Gaussian observation noise; The approximate expression of the non-Gaussian state noise is: ,in Non-Gaussian state noise An approximate representation of For approximation The number of Gaussian probability density functions, is the weight corresponding to the i-th component in the approximate representation of non-Gaussian state noise; The non-Gaussian observation noise is approximately expressed as: ,in Non-Gaussian state noise An approximate representation of For approximation The number of Gaussian probability density functions, represents the weight corresponding to the i-th component in the approximate representation of non-Gaussian observation noise; S14, obtaining an approximate representation of the posterior distribution based on the approximate representation of each non-Gaussian noise and the relationship between each non-Gaussian noise and the posterior distribution; In some embodiments, in step S14, obtaining an approximate representation of the posterior distribution based on the approximate representation of each non-Gaussian noise and the relationship between each non-Gaussian noise and the posterior distribution includes: The posterior distribution of the previous sampling point is approximated by a Gaussian probability density function group, which is expressed as: ,in represents the posterior distribution of the previous sampling point k-1, is the weight corresponding to the i-th component in the approximate representation of the posterior distribution of the previous sampling point; Substituting the approximate representation of non-Gaussian state noise into the state transfer model to approximate the state transfer model is expressed as: ； Substituting the approximate representation of non-Gaussian observation noise into the observation model to obtain an approximation of the distribution of observations obtained by the observation model is expressed as: ； Substitute the approximation of the state transition model and the posterior distribution of the previous sampling point k-1 into the prior distribution The approximation of the prior distribution is obtained as: ； The approximation of the prior distribution and the approximation of the observed value distribution are substituted into the posterior distribution to obtain an approximate representation of the posterior distribution.

2. The multi-source photoelectric detection data fusion method according to claim 1, characterized in that: The method of taking the mean of each Gaussian component as an unbiased estimate and performing data fusion optimization through a subgradient fusion algorithm to obtain a fused estimate of the target state quantity includes: The fused estimated value is set to be the sum of the products of the fusion weight matrices corresponding to each unbiased estimated value and the unbiased estimated value, and the sum of the fusion weight matrices is satisfied to be a unit matrix; The optimization objective function is determined to minimize the mean square error of the fused estimate, and the initial mean square error is obtained based on the covariance matrix or cross-covariance matrix between the initial fusion weight matrix and each unbiased estimate: Calculating the sub-gradient of the objective function for each of the fusion weight matrices; Updating the fusion weight matrix based on the subgradient; Calculate a new mean square error based on the updated fusion weight matrix. If the new mean square error is less than the mean square error before this iteration, update the mean square error to the mean square error obtained in this iteration. Otherwise, retain the mean square error and fusion weight matrix before the update. Iteratively updating the fusion weight matrix until a preset condition is met to obtain a final fusion weight matrix; A final fusion estimate of the target state quantity is obtained based on the final fusion weight matrix and the corresponding unbiased estimate.

3. The multi-source photoelectric detection data fusion method according to claim 2 is characterized in that: The objective function is equivalently determined as: ， in represents the trace of the matrix, represents the fusion weight matrix of the i-th unbiased estimate, represents the covariance matrix of the i-th unbiased estimate, represents the cross-covariance matrix between the ith unbiased estimator and the jth unbiased estimator.

4. A multi-source photoelectric detection data fusion device, characterized in that: include: An inference module, which obtains a posterior distribution of a state quantity of a target through Bayesian inference based on a state transition model and an observation model, wherein the state transition model includes non-Gaussian state noise and the observation model includes non-Gaussian observation noise; An approximation module, which sets the number of Gaussian components of the Gaussian mixture model to the number of photoelectric detection devices in the multi-source photoelectric detection system, and approximates the posterior distribution by expanding the Gaussian mixture model to determine the mean and variance of each Gaussian component; A fusion module, which uses the mean of each Gaussian component as an unbiased estimate, performs data fusion optimization through a subgradient fusion algorithm, and obtains a fusion estimate of the target state quantity; In step S1, the posterior distribution of the state quantity of the target is obtained by Bayesian inference based on the state transition model and the observation model, including: S11. Establish the state transition model and observation model of the system, expressed as: , where k represents the sampling point, represents the target state at sampling point k, represents the state transition model at sampling point k+1, represents non-Gaussian state noise, represents the system observation of the target at sampling point k, represents the observation model at sampling point k, represents non-Gaussian observation noise, where sampling point k is also the kth time step; S12, obtaining the posterior distribution of the state quantity according to Bayesian inference based on the state transfer model and the observation model; In step S12, obtaining the posterior distribution of the state quantity according to Bayesian inference based on the state transition model and the observation model includes: Based on the posterior distribution of the previous sampling point and the state transition model, the prior distribution of the current sampling point is obtained by prediction, which is expressed as: , Represents the observation based on the last sampling point The prior distribution of represents the state quantity distribution at the sampling point k obtained by the state transfer model, Represents the posterior distribution of the state quantity at the previous sampling point k-1; The prior distribution of the current sampling point is updated based on the observation model to obtain the posterior distribution of the current sampling point, which is expressed as: ,in Indicates that the sampling point k is based on the target state quantity through the observation model The observed value distribution of Represents the posterior distribution at the current sampling point k; S13, performing weighted approximation on the non-Gaussian state noise and the non-Gaussian observation noise through a Gaussian probability density function group to obtain approximate representations of the non-Gaussian state noise and the non-Gaussian observation noise respectively, and only approximating through a Gaussian mixture model; In step S13, weighted approximation is performed on the non-Gaussian state noise and the non-Gaussian observation noise respectively through a Gaussian probability density function group to obtain approximate representations of the non-Gaussian state noise and the non-Gaussian observation noise, including: Construct a group of Gaussian probability density functions with m groups of means and variances, expressed as: , represents the group of Gaussian probability density functions, where represents the mean vector of the i-th component in the Gaussian probability density function group, represents the covariance matrix of the i-th component in the Gaussian probability density function group, and n represents the dimension of the vector x; Performing weighted summation on the components of the Gaussian probability density function group to obtain respective approximations of the non-Gaussian state noise and the non-Gaussian observation noise; The approximate expression of the non-Gaussian state noise is: ,in Non-Gaussian state noise An approximate representation of For approximation The number of Gaussian probability density functions, is the weight corresponding to the i-th component in the approximate representation of non-Gaussian state noise; The non-Gaussian observation noise is approximately expressed as: ,in Non-Gaussian state noise An approximate representation of For approximation The number of Gaussian probability density functions, represents the weight corresponding to the i-th component in the approximate representation of non-Gaussian observation noise; S14, obtaining an approximate representation of the posterior distribution based on the approximate representation of each non-Gaussian noise and the relationship between each non-Gaussian noise and the posterior distribution; In some embodiments, in step S14, obtaining an approximate representation of the posterior distribution based on the approximate representation of each non-Gaussian noise and the relationship between each non-Gaussian noise and the posterior distribution includes: The posterior distribution of the previous sampling point is approximated by a Gaussian probability density function group, which is expressed as: ,in represents the posterior distribution of the previous sampling point k-1, is the weight corresponding to the i-th component in the approximate representation of the posterior distribution of the previous sampling point; Substituting the approximate representation of non-Gaussian state noise into the state transfer model to approximate the state transfer model is expressed as: ； Substituting the approximate representation of non-Gaussian observation noise into the observation model to obtain an approximation of the distribution of observations obtained by the observation model is expressed as: ； Substitute the approximation of the state transition model and the posterior distribution of the previous sampling point k-1 into the prior distribution The approximation of the prior distribution is obtained as: ； The approximation of the prior distribution and the approximation of the observed value distribution are substituted into the posterior distribution to obtain an approximate representation of the posterior distribution.

5. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the multi-source photoelectric detection data fusion method as described in any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the multi-source photoelectric detection data fusion method as described in any one of claims 1 to 3 is implemented.

Images