A method for classifying space objects based on physical behavior characteristics
By using a spatial target classification method based on physical behavior characteristics, and leveraging principal component analysis and Bayesian filtering output from feature samples, the problem of classification errors in traditional methods is solved, achieving accurate classification and real-time updated identification of spatial targets.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING INST OF CONTROL ENG
- Filing Date
- 2022-08-15
- Publication Date
- 2026-05-22
AI Technical Summary
Traditional space target classification methods are based on physical characteristics, which makes it difficult to accurately classify space targets that have multiple functions or are camouflaged, leading to classification errors.
A spatial target classification method based on physical behavior features is adopted. Through principal component analysis of feature samples, feature vector construction, preliminary classification of physical features, matching of behavioral features and Bayesian filtering output, accurate classification of spatial target types can be achieved, and the recognition accuracy can be updated in real time on the orbit.
It achieves accurate classification of space targets, is correctable and learnable, and can continuously update the identification accuracy based on the on-orbit identification situation, adapting to complex space situations.
Smart Images
Figure CN115600104B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a spatial target classification method based on physical behavior characteristics, belonging to the field of space satellite physical characteristics research. Background Technology
[0002] The characteristics of space targets can be divided into physical characteristics and behavioral characteristics. Physical characteristics include photometric characteristics, infrared characteristics, structural characteristics, and radar RCS characteristics, while behavioral characteristics include satellite orbital characteristics, relative velocity characteristics, angular velocity characteristics, and payload action characteristics. Space targets are diverse, and their characteristics vary greatly. Based on these different characteristics, space targets can be classified to determine satellite types, such as remote sensing satellites, reconnaissance satellites, communication satellites, and combat satellites. Traditional classification methods are mostly based on prior knowledge of physical characteristics, directly calculating classifications based on the payloads and configurations of specific satellite types. However, with the increasing complexity of the space situation and the continuous development of software-defined payloads, spacecraft are increasingly incorporating other functions or hiding their true functions beneath their surface features, such as remote sensing satellite platforms carrying reconnaissance payloads or communication satellite platforms carrying electromagnetic interference payloads. For such space targets, traditional classification methods may lead to classification errors. Therefore, it is necessary to establish a space target classification method that can comprehensively consider both physical and behavioral characteristics, and adjust and correct in real time on orbit. Summary of the Invention
[0003] The technical problem solved by this invention is to overcome the shortcomings of existing methods and propose a spatial target classification method based on physical behavior features. The method includes principal component analysis of feature samples and construction of classification feature vectors, preliminary classification of physical features and matching of behavioral features, and Bayesian filtering output of spatial target classification results. This method achieves effective and accurate classification of spatial target types, has correctability and learnability in the spatial target classification process, and can continuously update the recognition accuracy based on the on-orbit recognition situation, making it highly practical.
[0004] The objective of this invention is achieved through the following technical solutions:
[0005] A spatial target classification method based on physical behavior features includes:
[0006] A training dataset for typical target model features was constructed based on historical data of space targets measured in orbit.
[0007] Collect physical characteristic parameters of space targets to obtain characteristic sample information of space targets;
[0008] Principal component analysis is performed on the spatial target feature sample information to establish a feature vector sequence; the feature vector sequence consists of a physical feature vector sequence and a behavioral feature vector sequence.
[0009] Pre-judgment is performed on the spatial target type based on the physical feature vector sequence, and spatial target classification models are constructed according to the selected spatial target types.
[0010] The spatial target classification model is trained using the training dataset with the features of the typical target model.
[0011] Using the behavioral feature vector sequence as input to the spatial target classification model, the likelihood probability of the spatial target type is calculated, and the spatial target type with the maximum likelihood probability value is output to obtain the spatial target type sequence.
[0012] The sequence of space target types is subjected to Bayesian filtering, and the filtered space target types are output as the space target types.
[0013] In the above spatial target classification method, the expression for constructing spatial target feature sample information is:
[0014] S q =[O1(t),O2(t),...,O n (t),D1(t),D2(t),...,D m (t)]
[0015] Wherein, O1~On are the physical characteristic parameters of the space target, D1~Dm are the derived parameters generated based on the physical characteristic parameters of the space target, and t is the sampling time of the physical and behavioral characteristics.
[0016] In the above-mentioned space target classification method, the derived parameters D1 to Dm are obtained by performing second-order linear differentiation on the physical characteristic parameters of the space target. The formula for the second-order linear differentiation is:
[0017]
[0018] In the formula, ω(t) is the physical characteristic parameter of the space target, Ts is the sampling time, and y(t) is the derived parameter.
[0019] In the above spatial target classification method, the specific method for performing principal component analysis on the spatial target feature sample information to establish the feature vector sequence is as follows: Data normalization is performed on the spatial target feature sample information Sq to eliminate dimensional biases between different feature data; principal component factor analysis is performed on each target feature vector in the spatial target feature sample information Sq to identify key target features; and the top k target feature vectors whose sum of contributions to target classification exceeds 95% are selected to form the feature vector sequence:
[0020] Sq' = [m1, ..., mk]
[0021] In the formula, m1~mk are the target feature vectors selected from the spatial target feature sample information Sq.
[0022] In the above spatial target classification method, the spatial target type is pre-judged based on the physical feature vector sequence, and a spatial target classification model is constructed according to the selected spatial target type. The specific steps are as follows:
[0023] Step S41: Using a rule-based judgment method or a supervised learning method, based on the physical feature vector sequence in the feature vector sequence Sq', a preliminary classification of the target type is performed to obtain a target type set {λ1, λ2, ..., λn}, where each λ i Represents a type of space target, with each λ i Defined as a hidden Markov model describing the behavioral characteristics of a certain type of spatial target, the formula is:
[0024]
[0025] In the formula, Z is the set of all possible hidden states in the entire hidden Markov model, N is the number of hidden states, M is the set of all observable system states in the entire hidden Markov spatial target classification model, i.e., the set of system states of the feature vector sequence Sq', and K is the number of dominant states; π is the state probability distribution of the first set of data {m1(1), ..., mk(1)} of the feature vector sequence Sq', and the sum of all elements in π is 1; A is the hidden state probability transition matrix, and B is the emission probability matrix of the hidden Markov spatial target classification model; a ij For each element in matrix A, P(z) j |z i ) represents the hidden state Z i The state transitions to the hidden state Z. j The probability, b jk For each element in matrix B, P(m) k |z j The system exhibits observable m under the condition of hidden state Zj. k The probability of a state, m k For the observable state of the system, z N For the hidden state of the system, π N The probability corresponding to the initial state of the system;
[0026] Step S42: Using the typical target model feature training dataset and the Baum-Welch algorithm, iteratively calculate the state probability distribution π, the hidden state probability transition matrix A, and the emission probability matrix B. Among them, B is calculated using a Gaussian mixture model, and π and A are calculated using the Lagrange method. The training process of the hidden Markov model ends when the parameters π, A, and B converge.
[0027] In the above spatial target classification method, the step of using the behavioral feature vector sequence as input to the spatial target classification model, calculating the likelihood probability of the spatial target type, and outputting the spatial target type with the maximum likelihood probability value is specifically as follows: The feature vector sequence Sq' is used as input X to the hidden Markov model, and the likelihood probability of each λ in the target type set {λ1, λ2, ..., λn} is calculated. i The corresponding space target type likelihood probability P(X|λ) i ), calculate the maximum probability P of the feature vector sequence Sq'. max (X|λ i ), and output its corresponding space target type λ i This yields the spatial target type sequence Q = {λm, λn, ...}.
[0028] In the above-mentioned spatial target classification method, the step of performing Bayesian filtering on the spatial target type sequence and outputting the filtered spatial target type as the spatial target type is as follows:
[0029] Construct the input vector sequence of the Bayesian filter based on the spatial target type sequence, and design the Bayesian filter;
[0030] The expected probability of each spatial target type appearing in the input vector sequence of the Bayesian filter is calculated using the Bayesian filter, and a uniform detection threshold is set.
[0031] When the expected probability of the space target type is greater than the detection threshold, the corresponding space target type is output.
[0032] In the above-mentioned spatial target classification method, the specific method for constructing the input vector sequence of the Bayesian filter is as follows: Each element λ of the spatial target type sequence Q is... i This is represented by an L×1 vector x, where L represents the total number of space target types. Each value in vector x represents a type of satellite, and each value in x is either 0 or 1, with 1 indicating the corresponding element λ in Q. i For targets belonging to this spatial target type, the input vector sequence of the Bayesian filter is obtained as X = {x1, x2, x3, ..., xN}, where N is the number of samples.
[0033] In the above spatial target classification method, the design formula of the Bayesian filter is as follows:
[0034]
[0035] In the formula, p k n is the k-th element n in the input vector sequence X of the Bayesian filter. k =1 is the probability of occurrence, K is the number of states, K = L; n k Let k be the number of times the k-th state occurs during N trials. Let N be the distribution vector of the frequency of occurrence of each target type in N samples. The parameters are the multinomial probability distribution parameters of the target type. Let be the likelihood probability of the Bayesian filter, where
[0036] In the above spatial target classification method, the expected probability of each spatial target type appearing in the input vector sequence of the Bayesian filter is calculated using the Bayesian filter. The specific steps are as follows:
[0037] Step S71: Calculate the posterior probability using Bayes' theorem. The formula is as follows:
[0038]
[0039] In the formula, For posterior probability, For prior probability, Let P(X) be the likelihood probability, and let X = {x1, x2, x3, ..., xn} be the probability of the vector sequence X occurring. N Let} be the input vector sequence, where N is the number of samples; Where, p k n is the k-th element n in the input vector sequence X of the Bayesian filter. k The probability of 1 occurring;
[0040] Step S72: Calculate the prior probability mentioned in step S71 using the Dirichlet distribution formula. The formula is:
[0041]
[0042] In the formula, Γ() is the gamma function. Let be the parameters of the Dirichlet distribution, where
[0043] Step S73: Obtain the result of step S72 Substituting into step S71, the posterior probability expression in step S71 can be expressed as:
[0044]
[0045] in, Let be the parameters of the Dirichlet distribution, where n k Let X = {x1, x2, x3, ..., x} be the number of times the k-th state occurs during N trials. N Let} be the input vector sequence, where N is the number of samples; p k n is the k-th element n in the input vector sequence X of the Bayesian filter. k =1 is the probability of occurrence.
[0046] Step S74: Calculate the multinomial distribution parameters The expected value is given by the formula:
[0047]
[0048] In the formula, N is the number of consecutive samples of the Bayesian filter input vector, and X = {x1, x2, x3, ..., x N} represents the input vector sequence. Let n be the distribution vector of the frequency of occurrence of each target type during N sampling processes. k p represents the number of times the k-th state occurs during N trials. k n is the k-th element n in the input vector sequence X of the Bayesian filter. k The probability of 1 occurring. The initial value of αk is determined manually. Each time a filtering calculation is performed, the parameter α... k According to α k =α k +n k Iterative updates will be performed.
[0049] In the above spatial target classification method, when the expected probability of the spatial target type is greater than the detection threshold, the corresponding spatial target type is output. Specifically, the method involves setting an expected probability detection threshold P, and when a certain expected probability E(p) is calculated... k When the output result is greater than P, the target type corresponding to pk is output, which is the spatial target type of the Bayesian filter output.
[0050] The advantages of this invention over the prior art are as follows:
[0051] (1) The spatial target classification method based on physical behavior characteristics proposed in this invention integrates the behavioral characteristics of spatial targets on the basis of pre-classification using traditional classification methods. It explores the additional or hidden functions of spatial targets based on their behavior, and accurately classifies the behavior of spatial targets. The spatial target classification process is correctable and learnable.
[0052] (2) In the target classification process, if the input features of a certain target do not match the target category in the training model multiple times, a new spacecraft type can be set manually. When the input features reappear next time, it will be easier to classify them as the new target type.
[0053] (3) The present invention supplements the physical feature pre-classification based on the behavioral characteristics of space targets, and can continuously update the identification accuracy according to the on-orbit identification situation, which has strong practicality. Attached Figure Description
[0054] Figure 1 This is a schematic diagram of the spatial target Markov classification process provided in an embodiment of the present invention;
[0055] Figure 2 This is a schematic diagram of the Bayesian filtering iterative update process provided in an embodiment of the present invention. Detailed Implementation
[0056] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0057] like Figure 1 The diagram shows a schematic of the Markov classification process for spatial targets provided in an embodiment of the present invention. The present invention provides a spatial target classification method based on physical behavior characteristics, comprising the following steps:
[0058] Step (1): Construct a typical target model feature training dataset based on historical data of space targets measured in orbit;
[0059] Step (2): Collect the physical characteristic parameters of the space target to obtain the space target feature sample information, and construct the expression of the space target feature sample information as follows:
[0060] S q =[O1(t),O2(t),...,On (t),D1(t),D2(t),...,D m (t)]
[0061] Where O1~On are the physical characteristic parameters of the space target, D1~Dm are the derived parameters generated based on the physical characteristic parameters of the space target, and t is the sampling time of the physical and behavioral characteristics.
[0062] The derived parameters D1 to Dm are obtained by performing second-order linear differentiation on the physical characteristic parameters of the space target. The formula for the second-order linear differentiation is:
[0063]
[0064] In the formula, ω(t) is the physical characteristic parameter of the space target, Ts is the sampling time, and y(t) is the derived parameter.
[0065] Step (3): Perform principal component analysis on the spatial target feature sample information to establish a feature vector sequence; the feature vector sequence consists of a physical feature vector sequence and a behavioral feature vector sequence.
[0066] The specific method for establishing a feature vector sequence by performing principal component analysis on spatial target feature sample information is as follows: Data normalization is performed on the spatial target feature sample information Sq to eliminate dimensional biases between different feature data. Principal component factor analysis is then performed on each target feature vector in the spatial target feature sample information Sq to identify key target features. The top k target feature vectors, whose sum of contributions to target classification exceeds 95%, are selected to form the feature vector sequence:
[0067] Sq' = [m1, ..., mk]
[0068] In the formula, m1~mk are target feature vectors selected from the spatial target feature sample information Sq.
[0069] Step (4): Pre-judgment is performed on the spatial target type based on the physical feature vector sequence, and spatial target classification models are constructed according to the selected spatial target types. The specific steps are as follows:
[0070] Step S41: Using a rule-based judgment method or a supervised learning method, based on the physical feature vector sequence in the feature vector sequence Sq', a preliminary classification of the target type is performed to obtain a target type set {λ1, λ2, ..., λn}, where each λ i Represents a type of space target, with each λ i Defined as a hidden Markov model describing the behavioral characteristics of a certain type of spatial target, the formula is:
[0071]
[0072] In the formula, Z is the set of all possible hidden states in the entire hidden Markov model, N is the number of hidden states, M is the set of all observable system states in the entire hidden Markov spatial target classification model, i.e., the set of system states of the feature vector sequence Sq', and K is the number of dominant states; π is the state probability distribution of the first set of data {m1(1), ..., mk(1)} of the feature vector sequence Sq', and the sum of all elements in π is 1; A is the hidden state probability transition matrix, and B is the emission probability matrix of the hidden Markov spatial target classification model; a ij For each element in matrix A, P(z) j |z i ) represents the hidden state Z i The state transitions to the hidden state Z. j The probability, b jk For each element in matrix B, P(m) k |z j The system exhibits observable m under the condition of hidden state Zj. k The probability of a state, m k For the observable state of the system, z N For the hidden state of the system, π N The probability corresponding to the initial state of the system;
[0073] Step S42: Using the typical target model feature training dataset and the Baum-Welch algorithm, iteratively calculate the state probability distribution π, the hidden state probability transition matrix A, and the emission probability matrix B. Among them, the Gaussian mixture model is used to calculate B, and the Lagrange method is used to calculate π and A. The training process of the hidden Markov model ends when the parameters π, A, and B converge.
[0074] Step (5): Train the spatial target classification model using the training dataset of typical target model features;
[0075] Step (6): Using the behavioral feature vector sequence as input to the spatial target classification model, calculate the likelihood probability of the spatial target type, and output the spatial target type with the largest likelihood probability value to obtain the spatial target type sequence. The specific method is as follows: using the feature vector sequence Sq' as input X of the hidden Markov model, calculate the likelihood probability of each λ in the target type set {λ1, λ2, ..., λn}. i The corresponding space target type likelihood probability P(X|λ) i ), calculate the maximum probability P of the feature vector sequence Sq'. max (X|λ i ), and output its corresponding space target type λ i This yields the spatial target type sequence Q = {λm, λn, ...}.
[0076] Step (7): Perform Bayesian filtering on the spatial target type sequence, and output the filtered spatial target type as the spatial target type. The specific method is as follows:
[0077] (1) Construct the input vector sequence of the Bayesian filter based on the spatial target type sequence, and design the Bayesian filter;
[0078] The specific method for constructing the input vector sequence of a Bayesian filter is as follows: Each element λ of the spatial target type sequence Q... i This is represented by an L×1 vector x, where L represents the total number of space target types. Each value in vector x represents a type of satellite, and each value in x is either 0 or 1, with 1 indicating the corresponding element λ in Q. i For targets belonging to this spatial target type, at each time step, a Bayesian filter input vector can be obtained based on the target feature vectors input at this time step and the previous time step. By continuously sampling the Bayesian filter input vectors at N consecutive time steps, the Bayesian filter input vector sequence X = {x1, x2, x3, ..., xN} can be obtained, where N is the number of consecutive samples of the Bayesian filter input vector.
[0079] The design formula for a Bayesian filter is as follows:
[0080]
[0081] In the formula, p k n is the k-th element of the input vector sequence X of the Bayesian filter. k =1 is the probability of occurrence, that is, the probability that the target to be classified belongs to a certain type of spatial target; K is the number of states, that is, the number of possible target types, K = L; n k Let k be the number of times the k-th state occurs during N trials. Let N be the distribution vector of the frequency of occurrence of each target type in N samples. The parameters are the multinomial probability distribution parameters of the target type. Let be the likelihood probability of the Bayesian filter, that is, the prior probability. The probability of the input vector sequence X of a Bayesian filter of length N occurring is given by:
[0082] (2) Calculate the expected probability of each spatial target type appearing in the input vector sequence of the Bayesian filter using a Bayesian filter, and set a uniform detection threshold. The specific steps are as follows:
[0083] Step S71: Calculate the posterior probability using Bayes' theorem. The formula is as follows:
[0084]
[0085] In the formula, For posterior probability, For prior probability, Let P(X) be the likelihood probability, and let X = {x1, x2, x3, ..., xn} be the probability of the vector sequence X occurring. N Let} be the input vector sequence, where N is the number of samples; Where, p k n is the k-th element n in the input vector sequence X of the Bayesian filter. k The probability of 1 occurring;
[0086] Step S72: Calculate the prior probability of step S71 using the Dirichlet distribution formula. The formula is:
[0087]
[0088] In the formula, Γ() is the gamma function. Let be the parameters of the Dirichlet distribution, where
[0089] Step S73: Obtain the result of step S72 Substituting into step S71, the posterior probability expression in step S71 can be expressed as:
[0090]
[0091] in, Let be the parameters of the Dirichlet distribution, where n k Let X = {x1, x2, x3, ..., x} be the number of times the k-th state occurs during N trials. N Let} be the input vector sequence, where N is the number of samples; p k n is the k-th element n in the input vector sequence X of the Bayesian filter. k =1 is the probability of occurrence.
[0092] Step S74: Calculate the multinomial distribution parameters The expected value is given by the formula:
[0093]
[0094] In the formula, N is the number of consecutive samples of the Bayesian filter input vector, and X = {x1, x2, x3, ..., x N} represents the input vector sequence. Let n be the distribution vector of the frequency of occurrence of each target type during N sampling processes. kp represents the number of times the k-th state occurs during N trials. k n is the k-th element n in the input vector sequence X of the Bayesian filter. k The probability of 1 occurring. α k The initial value is determined manually. During the filtering process, the parameters are adjusted according to α for each filtering calculation. k =α k +n k Iterative updates in a manner such as Figure 2 As shown.
[0095] (3) When the expected probability of a space target type is greater than the detection threshold, the corresponding space target type is output. The specific method is as follows: set the expected probability detection threshold P, and when a certain expected probability E(p) is calculated, the corresponding space target type is output. k When the output result is greater than P, the target type corresponding to pk is output, which is the spatial target type of the Bayesian filter output.
[0096] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solutions of the present invention by utilizing the methods and techniques disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the protection scope of the technical solutions of the present invention.
Claims
1. A spatial target classification method based on physical behavior characteristics, characterized in that, include: Physical characteristics include photometric characteristics, infrared characteristics, structural characteristics, and radar RCS characteristics; Behavioral characteristics include satellite orbital characteristics, relative velocity characteristics, angular velocity characteristics, and payload action characteristics; A training dataset for typical target model features was constructed based on historical data of space targets measured in orbit. Collect physical characteristic parameters of space targets to obtain characteristic sample information of space targets; Principal component analysis is performed on the spatial target feature sample information to establish a feature vector sequence; The feature vector sequence consists of a physical feature vector sequence and a behavioral feature vector sequence; Pre-judgment is performed on the spatial target type based on the physical feature vector sequence, and spatial target classification models are constructed according to the selected spatial target types. The spatial target classification model is trained using the training dataset with the features of the typical target model. Using the behavioral feature vector sequence as input to the spatial target classification model, the likelihood probability of the spatial target type is calculated, and the spatial target type with the maximum likelihood probability value is output to obtain the spatial target type sequence. Perform Bayesian filtering on the spatial target type sequence and output the filtered spatial target type as the spatial target type; The method for performing Bayesian filtering on the spatial target type sequence and outputting the filtered spatial target type as the spatial target type is as follows: Construct the input vector sequence of the Bayesian filter based on the spatial target type sequence, and design the Bayesian filter; The expected probability of each spatial target type appearing in the input vector sequence of the Bayesian filter is calculated using the Bayesian filter, and a uniform detection threshold is set. When the expected probability of the space target type is greater than the detection threshold, the corresponding space target type is output. The spatial target type is pre-judged based on the physical feature vector sequence, and a spatial target classification model is constructed for each selected spatial target type. The specific steps are as follows: Step S41: Using a rule-based judgment method or a supervised learning method, based on the physical feature vector sequence in the feature vector sequence Sq', a preliminary classification of the target type is performed to obtain a target type set {λ1, λ2, ..., λ...}. n }, each Represents a type of space target, and will each Defined as a hidden Markov model describing the behavioral characteristics of a certain type of spatial target, the formula is: In the formula, Z is the set of all possible hidden states in the entire hidden Markov model, N is the number of hidden states, M is the set of all observable system states in the entire hidden Markov spatial object classification model, that is, the set of system states of the feature vector sequence Sq', and K is the number of dominant states. The first set of data {m1(1), ..., m} of the feature vector sequence Sq' k (1)} state probability distribution, The sum of all elements in the matrix is 1; A is the latent state probability transition matrix, and B is the emission probability matrix of the implicit Markov spatial target classification model. For each element in matrix A, For hidden state Z i The state transitions to the hidden state Z. j The probability, For each element in matrix B, For the system in hidden state Z j Under the condition of observable m k The probability of a state, m k For the observable state of the system, z N This is the hidden state of the system. The probability corresponding to the initial state of the system; Step S42: Using the typical target model feature training dataset and the Baum-Welch algorithm, calculate the state probability distribution. The hidden state probability transition matrix A and the emission probability matrix B are iteratively calculated. B is obtained using a Gaussian mixture model and a Lagrange multiplication method. and A, until the parameter When A, B and B all converge, the training process of the hidden Markov model ends.
2. The spatial target classification method based on physical behavior features according to claim 1, characterized in that: The expression for constructing spatial target feature sample information is: Among them, O1~ O n For the physical characteristic parameters of the space target, D1~D m These are derived parameters generated based on the physical characteristic parameters of the space target, where t is the sampling time of the physical and behavioral characteristics.
3. The spatial target classification method based on physical behavior features according to claim 2, characterized in that: The derived parameters D1~D m It is obtained by performing second-order linear differentiation on the physical characteristic parameters of the space target. The formula for the second-order linear differentiation is: In the formula, ω ( t ) represents the physical characteristic parameters of the space target. T s Sampling time, y ( t ) is the derived parameter.
4. The spatial target classification method based on physical behavior features according to claim 1, characterized in that: The specific method for performing principal component analysis on the spatial target feature sample information to establish the feature vector sequence is as follows: For the spatial target feature sample information S... q Data normalization is performed to eliminate dimensional biases between different feature data, and the spatial target feature sample information S is then processed. q Principal component factor analysis was performed on each target feature vector to identify key target features. The top k target feature vectors, whose sum of contributions to target classification exceeded 95%, were selected to form the feature vector sequence as follows: Sq’=[m1,…,m k ] In the formula, m1~m k To obtain the spatial target feature sample information S q The target feature vector selected from the list.
5. The spatial target classification method based on physical behavior features according to claim 1, characterized in that: The method of using the behavioral feature vector sequence as input to the spatial target classification model, calculating the spatial target type likelihood probability, and outputting the spatial target type with the maximum spatial target type likelihood probability is as follows: The feature vector sequence Sq' is used as input X to the hidden Markov model, and the target type set {λ1, λ2, ..., λ...} is calculated. n Each in} The corresponding spatial target type likelihood probability Calculate the maximum probability of the feature vector sequence Sq'. It also outputs the corresponding space target type. This yields the spatial target type sequence Q.
6. The spatial target classification method based on physical behavior features according to claim 1, characterized in that: The specific method for constructing the input vector sequence of the Bayesian filter is as follows: each element of the spatial target type sequence Q... This is represented by an L×1 vector x, where L represents the total number of space target types. Each value in vector x represents a type of satellite, and each value in x is either 0 or 1, with 1 indicating the corresponding element in Q. For targets belonging to this spatial target type, the input vector sequence of the Bayesian filter is obtained.
7. The spatial target classification method based on physical behavior features according to claim 1, characterized in that: When the expected probability of the space target type is greater than the detection threshold, the corresponding space target type is output. Specifically, a detection threshold P for the expected probability is set, and when a certain expected probability is calculated... When the output result is greater than P, output The corresponding target type is the spatial target type output by the Bayesian filter; The probability of an element occurring in the input vector sequence of the Bayesian filter.