A target stratification identification method based on dynamic combination weight sequential inspection and application

Abstract

This invention discloses a target hierarchical identification method and its application based on dynamic combined weight sequential test, comprising: establishing a feature model library; calculating the cost of each maneuvering stage of each target type using a dynamic combined weight method to obtain the maneuvering stage with the highest probability of the assumed target; performing dynamic combined weight calculation again on the targets after screening maneuvering stages to obtain the cost of the target belonging to each type, and converting it into the likelihood of belonging to each type; using a multi-hypothesis sequential probability ratio test to calculate the posterior probability of target identification, and determining the target type and termination condition; and updating the feature model library using the target identification results up to the current time. This invention can accurately identify targets with multiple maneuvering stages, multiple motion features, and overlapping motion features between the maneuvering stages of different targets in a short time.

Description

Technical Field

[0001] This invention relates to target hierarchical recognition technology, and in particular to a target hierarchical recognition method for targets with multiple ordered maneuvering stages, multiple motion characteristics, and overlapping motion characteristics between different target maneuvering stages. Background Technology

[0002] Target recognition is a crucial supporting technology, and efficient and accurate target recognition results can improve the accuracy of target tracking. Target recognition technology has been widely applied and developed in fields such as military reconnaissance, biological detection, drone control, and air traffic, and has significant research value.

[0003] Target recognition technology utilizes sensors to analyze detected target data and, combined with existing knowledge, accurately identifies the target type. Currently, target recognition methods mainly include image processing techniques using target image information and recognition based on target motion characteristics. Among these, algorithms using image processing techniques for target recognition have high computational complexity, demanding requirements on processing equipment, and are susceptible to interference from sensor measurements in harsh environments or when targets are highly maneuverable. On the other hand, existing algorithms using target motion characteristics only consider information from a single moment, failing to integrate historical data, and can only perform coarse classifications of targets. Furthermore, their accuracy is low when a single target exhibits multiple maneuvering phases.

[0004] The above difficulties can be summarized as the problem of target type identification under multiple maneuvering stages and strong maneuvering conditions, that is, it is necessary to make real-time judgments on the specific type of the target to be identified and the maneuvering stage. However, traditional target identification technology based on target motion characteristics has the following problems: (1) it does not consider the situation that the target has multiple maneuvering stages, the maneuvering stages are ordered, and the target motion characteristics differ greatly between each maneuvering stage; (2) it only considers the motion information at a single moment, and when the judgment at a single moment is wrong, it cannot use the correct judgment results at historical moments to correct it.

[0005] This paper addresses the hierarchical target recognition problem within the framework of dynamic weighting and sequential probability ratio testing. The challenges of hierarchical target recognition lie in the division of recognition levels, the classification of maneuvering phases for different target types, the selection of motion features, the determination of the weight ratios for various motion features, and the reasonable judgment based on information from multiple time points. Since targets have multiple maneuvering phases, the motion features corresponding to different maneuvering phases for the same type of target can vary significantly, and some motion features are not clearly distinguishable across different target types. Therefore, using only one set of motion features to describe each target type often leads to incorrect recognition results. For targets in specific maneuvering scenarios where a particular motion feature is prominent, using the same weighting calculation method as in other cases may result in significant errors in the recognition results. Traditional recognition algorithms cannot correct errors in single-step recognition, which also leads to incorrect results.

[0006] Therefore, providing a target recognition algorithm for situations where a target has multiple ordered maneuvering phases, multiple motion characteristics, and overlapping motion characteristics between different targets' maneuvering phases has become an urgent technical problem to be solved. Summary of the Invention

[0007] To address the aforementioned deficiencies in existing technologies, the present invention aims to provide a dynamic weighting calculation method that can adjust the proportion of each motion feature in the target recognition algorithm in real time, thereby improving the accuracy of target recognition. This invention combines a multi-hypothesis sequential probability ratio test method, comprehensively considering historical recognition information, and obtains the target recognition result through online calculation. Furthermore, it performs algorithm termination condition judgment during each test, thus achieving accurate recognition in a shorter time. Ultimately, it achieves accurate recognition of targets with multiple ordered maneuvering stages, multiple motion features, and overlapping motion features between different target maneuvering stages.

[0008] The present invention is achieved through the following technical solution.

[0009] According to one aspect of the present invention, a target hierarchical identification method based on dynamic combined weight sequential test is provided, comprising the following steps:

[0010] Based on the full-stage motion characteristics of all targets to be identified, motion characteristic models for different maneuvering stages of each type of target are established, typical characteristic data are set, and a characteristic model library is obtained.

[0011] A dynamic combination weighting method based on entropy weight and analytic hierarchy process is used to construct the target maneuvering phase identification cost function. The motion data of the target to be identified at the current moment is received, and the cost of the target to be identified belonging to each type of target and each different maneuvering phase is calculated to obtain the maneuvering phase with the highest probability under the assumption that the target belongs to each type.

[0012] A cost function for identifying each type of target is constructed. After the screening maneuver stage, the target is again subjected to dynamic combination weight calculation based on entropy weight and analytic hierarchy process to obtain the cost of the target to be identified belonging to each type of target. The cost is then converted into the likelihood of belonging to each type of target.

[0013] Based on the likelihood information of each type of target at each time step, the posterior probability of the target to be identified as each type of target up to the current time is calculated. The multiple hypothesis sequential probability ratio test is used to identify the type of target and determine the termination condition.

[0014] The feature model library is updated using the target recognition results up to the current time.

[0015] Regarding the above technical solution, the present invention has a further preferred embodiment:

[0016] Preferably, the step of establishing motion characteristic models for different maneuvering stages of various target types, setting typical characteristic data, and obtaining a feature model library includes:

[0017] Determine the set of target types and their corresponding maneuver phases;

[0018] Based on the type set and the maneuver stage set, factors that prominently reflect the type are selected as the typical motion characteristics of the target.

[0019] Feature analysis was performed on different ordered maneuver stages of various target types, and typical motion feature data were set for each type to obtain an initialized feature model library.

[0020] Preferably, obtaining the assumption that the target belongs to the most likely maneuver phase under each type of condition includes:

[0021] Receive motion feature data of the target to be identified at time n;

[0022] The objective weights ω of each motion feature are obtained using the principle of entropy weight. 1 ;

[0023] The subjective weights ω of each motion feature are obtained using the analytic hierarchy process (AHP). 2 ;

[0024] By objective weight ω 1 and subjective weight ω 2 The combined weight ω of each motion feature is calculated, thereby calculating the cost of the target belonging to each type of target and each maneuvering stage, and selecting the maneuvering stage with the highest probability under the assumption that the target belongs to each type.

[0025] Furthermore, the calculation of the objective weights of each motion feature includes:

[0026] Define the correlation coefficient between the motion data of the target to be identified and the typical characteristic data of each maneuver phase;

[0027] Calculate the entropy value of the i-th motion feature discrimination index for the j-th target;

[0028] Normalize the entropy value corresponding to the i-th motion feature of the j-th target;

[0029] The complementary value is calculated for the entropy value corresponding to the i-th motion feature of the normalized j-th target to obtain the entropy weight coefficient of the i-th feature index of the j-th target, i.e., the objective weight.

[0030] Furthermore, the calculation of the subjective weights of each motion feature includes:

[0031] The Analytic Hierarchy Process (AHP) is used to evolve the qualitative description of the importance relationship between features in target identification into a quantitative description.

[0032] Analyze the relationships between the various features of target identification, establish a hierarchical structure for target identification, and divide the indicator system into a target layer and an indicator layer;

[0033] Establish multiple criteria for judging the relative importance of motion features to determine the relative importance of features contained in the target:

[0034] The judgment criteria are selected based on the motion characteristics of the target to be identified, and a pairwise comparison judgment matrix is ​​constructed. Indicators belonging to the same category in the same layer are compared pairwise using the judgment criteria.

[0035] Perform a consistency check. If the consistency check result is less than the threshold, the consistency check passes; otherwise, readjust the judgment matrix.

[0036] Calculate the relative weight of the compared element with respect to the selected judgment criterion, which is the subjective weight of the flight feature corresponding to the compared element.

[0037] Furthermore, the calculation of the cost of the target to be identified belonging to each type of target at each maneuver stage includes:

[0038] The combined weight is calculated using both objective and subjective weights.

[0039] Calculate the weighted sum of the proximity of the target to be identified to all feature indicators and the proximity of each type of target to each maneuver stage, that is, the cost of the target to be identified belonging to each type of target to each maneuver stage.

[0040] By obtaining the most likely maneuver phases for each type of target under the assumption that the target belongs to each type, the maneuver phases for each type of target can be selected.

[0041] Preferably, the cost function for the target to be identified belonging to each type of target is calculated, and the cost is converted into the likelihood of belonging to each type of target, including:

[0042] Calculate the correlation coefficient between the motion characteristic data of the target to be identified and the typical characteristic data of each type of target corresponding to the maneuvering phase after screening;

[0043] The entropy values ​​corresponding to each motion feature based on the entropy weight principle are obtained from the correlation coefficient, and the objective weight w is calculated. 1 ;

[0044] The subjective weights w of each motion feature are obtained using the analytic hierarchy process (AHP). 2 :

[0045] The combined weight w is calculated using both objective and subjective weights.

[0046] Calculate the weighted sum of the proximity of all feature indicators to each type of target, which is the cost of the target to be identified belonging to each type of target.

[0047] Calculate the likelihood that the target to be identified belongs to each type of target at time n.

[0048] Preferably, a multi-hypothesis sequential probability ratio test is used to identify the target type and determine the termination condition, including:

[0049] At some point, a hypothesis H is generated for the j-th type of target. j The j-th class of targets is the true type of the target to be identified;

[0050] In the multiple hypothesis sequential probability ratio test, the posterior probability is used as the test statistic:

[0051] Define a threshold determination function; perform a termination condition check. If the test ends, output the result at that moment, and the result will not change afterward. Otherwise, continue to the next moment of detection.

[0052] Preferably, the feature model library is updated using the target recognition results up to the current time, including:

[0053] The maneuver phase at time n in which the j-th type of target has the highest probability is defined as follows:

[0054] At the cutoff time n, calculate the number of the most likely maneuver phases for each target within a given time range;

[0055] The number of maneuvers in each phase is determined, and the feature model library is updated accordingly.

[0056] The present invention, by adopting the above technical solution, has the following beneficial effects:

[0057] This invention proposes a novel dynamic combination weight calculation method, which can adjust the proportion of each motion feature in the target recognition algorithm in real time, thereby improving the accuracy of target recognition. This invention also constructs a hierarchical target recognition algorithm framework, which accurately identifies the maneuvering stage and specific type of a target when there are significant differences in motion features among different types of targets at multiple maneuvering stages and when there is overlap between the motion features of different types of targets. Furthermore, this invention combines the multi-hypothesis sequential probability ratio test method, comprehensively considers the recognition information from historical moments, obtains the target recognition result through online calculation, and performs algorithm termination condition judgment at each test, thus achieving accurate recognition in a shorter time.

[0058] This invention can be applied to fields such as military reconnaissance, biological detection, drone control, and air traffic. It can accurately identify targets with multiple maneuvering phases, multiple motion characteristics, and overlapping motion characteristics between different target maneuvering phases. It solves problems such as identification errors caused by multiple maneuvering phases of the target and the inability to correct errors in single-step identification results. Attached Figure Description

[0059] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, do not constitute an undue limitation of the invention. In the drawings:

[0060] Figure 1 Here is a flowchart of a target hierarchical identification algorithm based on dynamic combined weight sequential verification.

[0061] Figures 2(a)-(b) are schematic diagrams of the simulation scenarios;

[0062] The real target in Figure 2(a) is a hypersonic vehicle, and the real target in Figure 2(b) is a TBM reentry maneuvering warhead.

[0063] Figures 3(a)-(d) show the simulation results of the target hierarchical identification algorithm with and without the use of multiple hypothesis sequential probability ratio testing in the simulation scenario;

[0064] in Figures 3(a)-3(b) The results are presented in a simulated scenario using a target hierarchical recognition algorithm employing multiple hypothesis sequential probability ratio testing. Figures 3(c)-3(d) The results are from a target recognition algorithm that uses dynamic weight combination only, without employing hierarchical recognition and sequential verification in a simulated scenario. Figure 3(a) and 3(c) The simulation results correspond to the scenario shown in Figure 2(a). Figure 3(b) and 3(d) Simulation results corresponding to the scenario shown in Figure 2(b);

[0065] Figures 4(a)-(b) show the comparison of recognition results of the target hierarchical recognition algorithm with and without the use of the multi-hypothesis sequential probability ratio test.

[0066] Figure 4(a) corresponds to the simulation results of the scenario shown in Figure 2(a), and Figure 4(b) corresponds to the simulation results of the scenario shown in Figure 2(b). Detailed Implementation

[0067] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention.

[0068] like Figure 1 As shown, this embodiment of the invention provides a target hierarchical identification algorithm based on dynamic combined weight sequential verification, including the following steps:

[0069] S101: Based on the full-stage flight characteristics of all targets to be identified, establish motion (flight) characteristic models for different maneuvering stages of each type of target, set typical characteristic data, and obtain a characteristic model library.

[0070] Specifically, the steps include the following:

[0071] 11) Determine the set of target types and the corresponding set of maneuver stages to be identified, and let the number of target types be M and the number of maneuver stages be S;

[0072] 12) Based on the type set and the maneuver phase set, select the factors that reflect the type prominently as the typical flight characteristics of the target, and let the number of flight characteristics be N;

[0073] 13) Perform feature analysis on different ordered flight phases of various target types, and set typical flight feature data for each type to obtain an initialized feature model library, denoted as x. Also, denot the typical flight feature data of the j-th type of target in the l-th flight phase as x. j,l ={x j,l (i)}, 1≤i≤N, 1≤j≤M, 1≤l≤S, where i represents the i-th flight feature.

[0074] S102: A target maneuvering phase identification cost function is constructed using a dynamic combination weighting method based on entropy weight and analytic hierarchy process. The flight data of the target to be identified at the current moment is received, and the cost of the target to be identified belonging to each type of target and each maneuvering phase is calculated to obtain the maneuvering phase with the highest probability under the assumption that the target belongs to each type.

[0075] Specifically, the steps include the following:

[0076] 21) Receive the motion characteristic data of the target to be identified at time n; let Let i be the flight data of the target to be identified at time n, where i represents the i-th flight feature;

[0077] 22) The objective weights ω of each flight characteristic are obtained using the principle of entropy weight. 1 ;

[0078] The calculation steps for objective weights are as follows:

[0079] 221) Define the correlation coefficient between the flight data of the target to be identified and the characteristic typical data of each maneuver phase, as follows:

[0080]

[0081] In the formula, q j (i,l) represents the correlation coefficient between the i-th flight characteristic data of the target to be identified and the typical characteristic data of the l-th maneuvering phase of the j-th type of target; x j,l (i) represents the typical flight feature data of the i-th flight feature of the j-th target in the l-th flight phase; μ is the resolution coefficient, used to adjust the size of the comparison environment, with a value range of [0,1]. When μ = 0, the comparison environment disappears, and when μ = 1, the influence of the comparison environment on the correlation coefficient reaches its maximum. Generally, the value is μ = 0.5; S is the number of maneuver phases.

[0082] 222) Calculate the entropy value of the i-th flight feature discrimination index for the j-th target:

[0083]

[0084] In the formula, E i,j This represents the entropy value corresponding to the i-th flight feature of the j-th target;

[0085] 223) For E i,j Normalize:

[0086]

[0087] In the formula, E max =ln S, where S is the number of maneuver phases;

[0088] 224) for e i,j Find the complementary value to obtain the entropy weight coefficient of the i-th feature index of the j-th target, i.e., the objective weight:

[0089]

[0090] In the formula, This represents the objective weight corresponding to the i-th flight feature of the j-th target;

[0091] 23) The subjective weights ω of each flight characteristic are obtained using the analytic hierarchy process (AHP). 2 ;

[0092] The calculation steps for subjective weights are as follows:

[0093] 231) The Analytic Hierarchy Process (AHP) is used to develop the relationship between the importance of various criteria in target identification from a qualitative description to a quantitative description;

[0094] 232) Establish a hierarchical structure for target identification. Analyze the relationships between various features of target identification, and based on the target maneuvering phase and type identification problems to be solved, consider specific target features and divide the indicator system into a target layer and an indicator layer: the maneuvering phase corresponding to each type of target is defined as the target layer; different maneuvering phases are represented by different typical feature data, and flight features form the indicator layer;

[0095] 233) Establish multiple criteria for judging the relative importance of flight features to determine the relative importance of features contained in the target:

[0096] First, establish the relative importance sequence of N flight features at the initial time;

[0097] Define N judgment threshold parameters corresponding to flight characteristic indicators, denoted as follows: 1≤i≤N, the value of ρ is related to the specific type and flight characteristics of the target;

[0098] Threshold judgment is performed on the flight data received at the current moment:

[0099] If flight data exists Make or This indicates that the feature is relatively prominent, and some types of targets that may be identified can be excluded by using this feature. Therefore, when comparing indicators in the future, the relative importance judgment criterion that makes this flight feature prominent should be selected.

[0100] 234) Construct pairwise comparison judgment matrices. Perform pairwise comparisons on indicators belonging to the same category within the same layer. For the j-th category of targets, denot the judgment matrix obtained from the pairwise comparisons as C. j :

[0101]

[0102] In the formula, c mn This indicates the relative importance of the m-th flight feature compared to the n-th flight feature;

[0103] 235) Consistency verification. Introducing consistency metrics:

[0104]

[0105] In the formula, λ max To determine the largest eigenvalue of matrix C; N is the number of flight features;

[0106] Calculate the consistency ratio:

[0107]

[0108] In the formula, RI is the average random consistency index; when CR < 0.10, the consistency check passes, otherwise the judgment matrix is ​​readjusted.

[0109] 236) Calculate the relative weight of the compared element with respect to the selected judgment criterion, which is the subjective weight of the flight feature corresponding to the compared element:

[0110]

[0111] In the formula, Let represent the subjective weight corresponding to the i-th flight feature of the j-th target, and ∑ be the summation symbol;

[0112] 24) By objective weight ω 1 and subjective weight ω 2 The combined weight ω of each flight characteristic is calculated, thereby calculating the cost of each maneuver phase of the target and enabling the selection of the target maneuver phase.

[0113] The calculation steps for the combined weights and costs are as follows:

[0114] 241) Calculate the combined weight using both objective and subjective weights. The formula for calculating the combined weight is as follows:

[0115]

[0116] In the formula, ω i,j This represents the weight corresponding to the i-th feature index of the j-th target;

[0117] 242) Calculate the weighted sum of the proximity of the target to be identified to all characteristic indicators and the proximity of each type of target to each maneuver stage, using A. j,l This represents the weighted sum of the proximity of the target to be identified to all feature indicators and the proximity of the target to class j in the l-th maneuver phase, i.e., the cost of the target to be identified belonging to class j in the l-th maneuver phase:

[0118]

[0119] For the j-th type of target, different maneuver stages correspond to different A values. j,l A j,l The smaller the value, the greater the probability that the target to be identified is in the l-th maneuvering stage of the j-th target category.

[0120] Under the assumption that the target to be identified belongs to the j-th target category, the most likely maneuver phase is denoted as . By identifying the most probable maneuvering phase for each type of target under the assumption that the target belongs to each category, the maneuvering phases for each type of target can be selected:

[0121]

[0122] S103: Construct cost functions for identifying various types of targets, and perform dynamic combination weight calculation based on entropy weight and analytic hierarchy process on the targets after the screening maneuver stage to obtain the cost of the target to be identified belonging to each type of target, and convert the cost into the likelihood of belonging to each type of target.

[0123] Specifically, the steps include the following:

[0124] 31) Calculate the correlation coefficient between the flight data of the target to be identified and the characteristic typical data of the corresponding maneuver phase of the screened target:

[0125]

[0126] In the formula, This represents the flight feature data of the i-th target to be identified and the data of the j-th target. The correlation coefficients between typical characteristic data of each maneuver phase; M is the number of target types;

[0127] 32) Based on the correlation coefficient The entropy values ​​corresponding to each flight feature based on the entropy weight principle are obtained:

[0128]

[0129] In the formula, ζ i This represents the entropy value corresponding to the i-th flight feature;

[0130] Then the objective weight w 1 It can be calculated using the following formula:

[0131]

[0132]

[0133] In the formula, ζ max =lnM; This represents the objective weight corresponding to the i-th flight feature;

[0134] 33) The subjective weights w of each flight characteristic are obtained using the analytic hierarchy process (AHP). 2 :

[0135] Each type of target is defined as the target layer, and the selected maneuver phases are represented by corresponding characteristic typical data, forming the index layer of flight characteristics;

[0136] Same as step 234), construct the judgment matrix D = [d mn ] N×N , where d mn This represents the relative importance of the m-th flight feature compared to the n-th flight feature; N is the number of flight features; and a consistency check is performed.

[0137] The subjective weight is calculated using the following formula:

[0138]

[0139] In the formula, This represents the subjective weight corresponding to the i-th flight feature;

[0140] 34) Calculate the portfolio weights:

[0141]

[0142] 35) For the target to be identified, use B j This represents the weighted sum of the proximity differences between all feature indices and the proximity differences between the target and the j-th class of targets, i.e., the cost of the target to be identified belonging to the j-th class of targets:

[0143]

[0144] In the formula, This represents the maneuver phase with the highest probability under the assumption that the target to be identified is of the j-th type.

[0145] 36) Define the likelihood function of the target to be identified at time n as belonging to the j-th class of targets as:

[0146]

[0147] S104: Based on the likelihood information of each type of target at each time step, calculate the posterior probability of the target to be identified as each type of target up to the current time step, and use the multiple hypothesis sequential probability ratio test to identify the type of target and determine the termination condition.

[0148] Specifically, the steps include the following:

[0149] 41) In order to reduce the impact of single-moment calculation errors on target recognition results and to calculate target recognition results online, a multi-hypothesis sequential probability ratio test is used to make decision-making and termination judgments based on the flight data of the target to be identified.

[0150] At time n, a hypothesis H is generated for the j-th type of target. j The j-th class of targets is the true type of the target to be identified;

[0151] In the multiple hypothesis sequential probability ratio test, for the target flight characteristic data up to time n, the following posterior probability is used as the test statistic:

[0152]

[0153]

[0154] In the formula, This can be obtained from Bayes' theorem:

[0155]

[0156] In the formula, π j This represents the prior probability that the target to be identified belongs to the j-th class of targets, and is generally taken as π. j =1 / M; Π is the product symbol, indicating the product of consecutive terms;

[0157] 42) Define the threshold determination function Π(p):

[0158]

[0159] In the formula, η is the threshold parameter;

[0160] Up to time n, if there is one and only target of type j that satisfies the following formula, then the target identification result can be considered as j, and it will not change further. Otherwise, the detection continues to the next time step:

[0161]

[0162] In the formula, T1 is the time parameter, and T1≥n; σ1 is the threshold parameter.

[0163] S105: Update the feature model library using the target recognition results up to the current time.

[0164] Specifically, the steps include the following:

[0165] 51) Let the phase at time n in which the probability of the j-th type of target's maneuver is highest be denoted as .

[0166] 52) Calculate the number of times the target to be identified is determined to be in the l-th maneuver phase for the j-th type of target within a given time range [n-T2+1,n] (T2≥n) at the cutoff time n:

[0167]

[0168]

[0169] 53) If at cutoff time n, there exists l′ such that the following equation holds, then for the j-th type of target, the feature model of the stage preceding this maneuver will be removed during the next identification, thereby updating the feature model library:

[0170] Q j,l′ >σ2T2

[0171] In the formula, σ2 is the threshold parameter.

[0172] The invention will be further illustrated below with specific examples.

[0173] S101: Based on the full-stage motion characteristics, i.e., flight characteristics, of all targets to be identified, the possible types of targets and their corresponding maneuvering phases are determined. There are two types of targets here: hypersonic vehicles and TBM reentry maneuvering warheads. Both maneuvering phases include a boost phase, a glide phase, and a reentry phase. Then, factors that prominently reflect the type are selected as typical flight characteristics of the target; here, flight altitude, flight speed, and maneuvering overload are selected. Feature analysis is performed on the different ordered flight phases of each target, and typical flight characteristic data are set for each, resulting in an initialized feature model library. The typical feature data corresponding to the two types of targets are shown in Table 1, where Ma is the Mach number and g is the gravitational acceleration.

[0174] Table 1 Typical flight characteristics data of the two types of targets

[0175]

[0176] S102: Construct a target maneuvering phase identification cost function. Receive the flight data of the target to be identified at the current moment, and use the entropy weight principle to obtain the objective weight ω for each flight feature belonging to each type of target and each maneuvering phase. 1 The subjective weights ω of each flight characteristic are obtained using the analytic hierarchy process (AHP). 2 , by objective weight ω 1 and subjective weight ω 2 The combined weight ω of each flight feature is calculated, thereby calculating the cost of the target to be identified belonging to each type of target and each maneuvering stage, and selecting the maneuvering stage with the highest probability under the assumption that the target belongs to each type.

[0177] S103: Construct cost functions for identifying various target types. For the flight data corresponding to each target type after the maneuvering phase, again use the entropy weight principle to obtain the objective weight w of each flight feature belonging to each target type. 1 The subjective weight w of each flight characteristic belonging to each type of target was obtained using the analytic hierarchy process. 2 , determined by objective weight w 1 and subjective weight w 2The combined weight w of each flight feature belonging to each type of target is calculated, then the cost of the target to be identified belonging to each type is calculated, and then it is transformed into the likelihood of belonging to each type.

[0178] S104: Based on the likelihood information of each type of target at each time step, a multi-hypothesis sequential probability ratio test is used to generate hypothesis H for the j-th type of target at time n. j The j-th class is the true type of the target to be identified; calculate the posterior probability p. n The threshold decision function Π(p) is defined and used as the test statistic. Then, the termination condition is judged. If the termination is completed, the result at that time is output and will not change afterward. Otherwise, the detection at the next time is continued.

[0179] S105: At the cutoff time n, calculate the number of maneuvers most likely for each type of target within the given time range; determine the number of each maneuver phase and update the feature model library.

[0180] Figures 2(a) and (b) show simulation scenarios for aerial target identification. The targets maneuver in three-dimensional space, all described using a geocentric Cartesian coordinate system. In the simulation scenario shown in Figure 2(a), the target is a hypersonic vehicle, undergoing three maneuvering phases: boost, glide, and reentry. The glide phase involves a longitudinal jump maneuver. In the simulation scenario shown in Figure 2(b), the target is a TBM reentry maneuvering warhead, undergoing three maneuvering phases: boost, glide, and reentry. The reentry phase involves a lateral maneuver. During target identification, the identification interval T = 1 s. The maneuvering time is 360 s in the scenario where the target is a hypersonic vehicle and 770 s in the scenario where the target is a hypersonic vehicle.

[0181] Figures 3(a)-(b) show the target recognition probabilities of the target hierarchical recognition algorithm based on dynamic combined weight sequential verification in the simulated scenarios. In (a), the true target category in the simulated scenario is a hypersonic aircraft, and in (b), the true target category in the simulated scenario is a reentry maneuvering warhead. It can be seen that for both types of targets, the algorithm proposed in this invention can correctly identify the target type in a relatively short time. Furthermore, due to the addition of a termination condition, the algorithm terminates after only 30 seconds, demonstrating high efficiency.

[0182] Figures 3(c)-(d) show the target recognition probabilities in the simulated scenarios where no hierarchical recognition and sequential verification were used, but only dynamic weighting was employed. In (c), the true target category in the simulated scenario was a hypersonic vehicle, and in (d), the true target category was a reentry vehicle. Although the probability of identifying a target correctly was higher than that of identifying a target incorrectly, the difference was small. Furthermore, when identifying hypersonic vehicles, identification errors occurred between 20 and 130 seconds and after 350 seconds.

[0183] Figures 4(a)-(b) show a comparison of the recognition results of the two algorithms. Here, a target is defined as being identified as belonging to a certain class of targets when the recognition probability is higher than 0.6; when the recognition probability is lower than 0.4, it is defined as being identified as belonging to another class of targets; and when the recognition probability is between 0.4 and 0.6, it is defined as unrecognizable. Recognition Algorithm 1 is the target hierarchical recognition algorithm based on dynamic combined weight sequential verification proposed in this invention; Recognition Algorithm 2 is a target recognition algorithm that does not employ hierarchical recognition and sequential verification, but only uses dynamic combined weights. It can be seen that the algorithm proposed in this invention reduces the recognition time, reduces the impact of single-step errors on the recognition results, and improves the stability and accuracy of the algorithm's recognition results.

[0184] The above experimental results verify that the present invention improves the accuracy of target recognition and achieves accurate recognition of targets with multiple ordered maneuvering stages, multiple motion characteristics, and overlapping motion characteristics between different target maneuvering stages, which has significant theoretical and practical value.

[0185] This invention is not limited to the above embodiments. Based on the technical solutions disclosed in this invention, those skilled in the art can make some substitutions and modifications to some of the technical features without creative effort, and all such substitutions and modifications are within the protection scope of this invention.

Claims

1. A target hierarchical identification method based on dynamic combined weight sequential verification, characterized in that, include: Based on the full-stage motion characteristics of all targets to be identified, motion characteristic models for different maneuvering stages of each type of target are established, typical characteristic data are set, and a characteristic model library is obtained. The motion characteristics are those of flight. A dynamic combination weighting method based on entropy weight and analytic hierarchy process is used to construct the target maneuvering phase identification cost function. The motion data of the target to be identified at the current moment is received, and the cost of the target to be identified belonging to each type of target and each different maneuvering phase is calculated to obtain the maneuvering phase with the highest probability under the assumption that the target belongs to each type. A cost function for identifying each type of target is constructed. After the screening maneuver stage, the target is again subjected to dynamic combination weight calculation based on entropy weight and analytic hierarchy process to obtain the cost of the target to be identified belonging to each type of target. The cost is then converted into the likelihood of belonging to each type of target. Based on the likelihood information of each type of target at each time step, the posterior probability of the target to be identified as each type of target up to the current time is calculated. The multiple hypothesis sequential probability ratio test is used to identify the type of target and determine the termination condition. The feature model library is updated using the target recognition results up to the current time.

2. The method according to claim 1, characterized in that, The process involves establishing motion characteristic models for different maneuvering stages of various target types, setting typical characteristic data, and obtaining a characteristic model library, including: Determine the set of target types and their corresponding maneuver phases; Based on the type set and the maneuver stage set, factors that prominently reflect the type are selected as the typical motion characteristics of the target. Feature analysis was performed on different ordered maneuver stages of various target types, and typical motion feature data were set for each type to obtain an initialized feature model library.

3. The method according to claim 1, characterized in that, The assumption that the target belongs to the most likely maneuver phase under each type of condition includes: take over Motion characteristic data of the target to be identified at any time; The objective weights of each motion feature are obtained by using the principle of entropy weight; The subjective weights of each motion feature are obtained using the analytic hierarchy process (AHP). The combined weights of each motion feature are calculated using objective and subjective weights. The cost of the target to be identified belonging to each type of target and each maneuvering stage is calculated, and the maneuvering stage with the highest probability under the assumption that the target belongs to each type is selected.

4. The method according to claim 3, characterized in that, The objective weight calculation for each motion feature includes: Define the correlation coefficient between the motion data of the target to be identified and the typical characteristic data of each maneuver phase; Calculate the first Class Target No. The entropy value of a motion feature discrimination index; For the The first class of targets Normalize the entropy values ​​corresponding to each motion feature; For the normalized first The first class of targets Calculate the complementary value of the entropy value corresponding to the i-th motion feature to obtain the i-th... Class Target No. The entropy weight coefficient of each feature indicator, i.e., the objective weight.

5. The method according to claim 3, characterized in that, The subjective weight calculation for each motion feature includes: The Analytic Hierarchy Process (AHP) is used to evolve the qualitative description of the importance relationship between features in target identification into a quantitative description. Analyze the relationships between the various features of target identification, establish a hierarchical structure for target identification, and divide the indicator system into a target layer and an indicator layer; Establish multiple criteria for judging the relative importance of motion features to determine the relative importance of features contained in the target: The judgment criteria are selected based on the motion characteristics of the target to be identified, and a pairwise comparison judgment matrix is ​​constructed. Indicators belonging to the same category in the same layer are compared pairwise using the judgment criteria. Perform a consistency check. If the consistency check result is less than the threshold, the consistency check passes; otherwise, readjust the judgment matrix. Calculate the relative weight of the compared element with respect to the selected judgment criterion, which is the subjective weight of the flight feature corresponding to the compared element.

6. The method according to claim 3, characterized in that, The calculation yields the cost of the target belonging to each type of target and each maneuvering phase, including: The combined weight is calculated using both objective and subjective weights. Calculate the weighted sum of the proximity of the target to be identified to all feature indicators and the proximity of each type of target to each maneuver stage, that is, the cost of the target to be identified belonging to each type of target to each maneuver stage. By obtaining the most likely maneuver phases for each type of target under the assumption that the target belongs to each type, the maneuver phases for each type of target can be selected.

7. The method according to claim 1, characterized in that, The cost of identifying a target belonging to each target type is calculated, and the cost is converted into the likelihood of belonging to each target type, including: Calculate the correlation coefficient between the motion data of the target to be identified and the characteristic typical data of the corresponding maneuver phase of the screened target; The entropy values ​​corresponding to each motion feature based on the entropy weight principle are obtained from the correlation coefficient, and the objective weight is calculated. The subjective weights of each motion feature are obtained using the analytic hierarchy process (AHP). The combined weight is calculated using both objective and subjective weights. Calculate the weighted sum of the proximity of all feature indicators to each type of target, which is the cost of the target to be identified belonging to each type of target. Calculate the target to be identified at time... Likelihoods for each type of objective.

8. The method according to claim 1, characterized in that, The multi-hypothesis sequential probability ratio test is used to identify the target type and determine the termination condition, including: At some point, for the first Class target generation hypothesis: The The class is the true type of the target to be identified; In the multiple hypothesis sequential probability ratio test, the posterior probability is used as the test statistic: Define a threshold determination function; perform a termination condition check. If the test ends, output the result at that moment, and the result will not change afterward. Otherwise, continue to the next moment of detection.

9. The method according to claim 1, characterized in that, The feature model library is updated using the target recognition results up to the current time, including: Set a certain time. The phase in which the target type is most likely to maneuver; Up to a certain point in time, calculate the number of times the target to be identified has the highest probability of maneuvering phase for each target within a given time range; The number of maneuvers in each phase is determined, and the feature model library is updated accordingly.

10. The method according to any one of claims 1-9 is applied to the accurate identification of targets in military reconnaissance, biological detection, unmanned aerial vehicle control, and air traffic.

Images